DER GROSSE KI-BREMSSCHWINDEL

00:00:00: Let me be clear.

00:00:02: A super intelligent AI that escapes human control will not an American problem, it won't a Chinese problem but humanity's problems.

00:00:16: The future of humanity cannot left in the hands of a handful of big tech oligarchs.

00:00:29: Thank you very much.

00:00:33: And this time we were also about to say hello and talk about fast, quick

00:00:50: developments while everyone actually agreed on the handbrake.

00:00:57: But well... it doesn't seem like

00:01:00: all of us, right?

00:01:02: Yes!

00:01:03: In that sense welcome to Wunder Panic your podcast.

00:01:06: between anger and stress.

00:01:08: We are Kata and Milady who is back today from A.I.tonomy.

00:01:13: By the

00:01:14: way here you just became a starting fan.

00:01:19: Super introduction today.

00:01:20: Attention,

00:01:22: here is a quote from the White House.

00:01:25: considers anyone that uses the term artificial intelligence as opposed to the highly accepted new and more accurate term superintelligence

00:01:36: The

00:01:37: Enemy?

00:01:37: Thank you very much for your attention, President Donald Duck Trump

00:01:44: I would say we have made it, Katta We are proud of Donald Trump

00:01:48: Exactly.

00:01:48: He wrote in a true social post, I think two or three days ago we have the tenth of October today.

00:01:57: he wrote there that the term Artificial Intelligence does not count anymore as how it is now called super intelligence.

00:02:07: and all those who refuse to say this are the enemies.

00:02:16: It's totally surreal.

00:02:18: I read an article in which he confirmed once again that the Chinese president is also a good fan, if you call it that way.

00:02:26: So they're already too far away and think only to be too far today because when we look at this term used now ... not everyone does!

00:02:39: Meta and Elon Musk said, okay, everything is clear.

00:02:42: Mark Zuckerberg and Elon musk are already there and also use super intelligence but as much I know both Enthropic as well as Open AI.

00:02:51: they still say artificial intelligence.

00:02:53: we just do that now or because We're so revolutionary And stay with me.

00:02:59: in

00:02:59: any case i will simply stay at KI or in large language models, because it's enough actually.

00:03:10: It is enough with over-ordered terms that no one can order anymore ... What

00:03:17: are we doing today?

00:03:18: We have to get back together again!

00:03:19: Yes finally back

00:03:21: together!

00:03:21: Finally back together at a Samstag

00:03:23: this time!

00:03:23: I liked the fact that we're here again.

00:03:25: We recently personally met each other and that was also totally beautiful but then we didn't even come up on the podcast.

00:03:30: That's why it is so much nicer to sit together like a microphone.

00:03:37: Absolutely,

00:03:38: we ate and went shopping!

00:03:40: And we worked

00:03:43: on the song and were

00:03:44: working on the neosome as well Yes, and somehow a lot happened.

00:03:47: And we thought that today would be the day where we'll do another recap of our last few weeks as well.

00:03:55: Let's take a quick look at what actually went wrong because it is not just like President of the United States says we call this super-intelligence.

00:04:03: There are still some things going on around here.

00:04:05: We're looking forward to seeing you again!

00:04:09: We'll take you with us in the madness... To my disappointment,

00:04:14: I have to say that on a huge Miroboard with lots of notes, I'm going crazy.

00:04:22: Because it wasn't just summer holidays and we were on vacation... No!

00:04:26: We also had an amazing pipeline together as guests in the audience.

00:04:29: They

00:04:30: were all great too.

00:04:30: Yes they are.

00:04:31: I think this is even our birthday episode because the last one was in June when we sat down alone.

00:04:39: And...

00:04:39: Bad.

00:04:40: That was great, but it's bad that we're so many.

00:04:42: Exactly and because of this... We thought today to look at our cars a little bit

00:04:44: again in order to see what happened.

00:04:50: So I would suggest breaking down once What has been since the last episode alone on June thirteenth Because i think all of us are like this It constantly sweeps something over the pond Over to us.

00:05:10: There were big headlines, but there was also repeatable stories.

00:05:17: I think it all happened at least over the summer and now in recent weeks... ...but also days and hours.

00:05:24: And what is actually still right?

00:05:29: What's going on here?

00:05:29: What does that mean for you guys today?

00:05:33: That's why I would say let me just start off like this.

00:05:39: in what actually happens.

00:05:40: And some of them we take it easy, everything that I'm telling you about is still my show notes.

00:05:51: but now the timeline comes to an end.

00:05:55: We'll start a bit earlier on the first and eighth of June because there will definitely be Enthropic and then OpenAI, trust me, the evil plans at SEC.

00:06:08: That means

00:06:09: that if you want to go on the drill, then you have to make a plan beforehand.

00:06:13: We have it before we do this.

00:06:15: Okay

00:06:15: and shortly later at the twelfth and sixth, which was also in our birthday episode , Space X is going there so including xai mask stands behind him with his company goes to the drill and that's the biggest drill ever in history.

00:06:29: seventy-five million years ago short on or in the time span between the twelfth sixth and the first seventh there is then we remember maybe all of us export controls anthropic sets access to fabel five and mythos five out after that was basically forbidden.

00:06:53: and Then again, there's a lift up and they set it free again.

00:06:58: Then there were twenty-eighth sevenths.

00:07:00: I think We've been waiting for this so much.

00:07:03: There was an open letter.

00:07:05: It's Pacing the Frontier from over a thousand and one hundred employees of different KILAPS, so they basically demanded please, please do it slowly again around ten o'clock.

00:07:20: So we were just both somehow let go of the saw in Ulob.

00:07:24: Then Sanderz, whom you have heard at the beginning of a game, called Sam Altman and Dario Amadei.

00:07:33: Once the CEO of Open AI and once the one from Anthropik And also Zuckerberg who stands behind Meta So everything that he knows.

00:07:42: so under Facebook and Instagram then asked him to stop developing.

00:07:45: No ten days later Then we get an open AI report.

00:07:52: Up to seven hundred own agents have then coordinated a hack via a self-made news board that connects with this hugging face case.

00:08:06: We'll talk about it in detail again shortly later, about ten days later on the ninth, there comes a very unknown person Jacob Coxon of X and publishes by Anthropic, writes the industry with our lives.

00:08:25: It received more than one hundred fifty million calls.

00:08:30: And then even twenty or even more congress members appear immediately in the USA after regulation.

00:08:40: On the twelfth of the ninth day – so you can see that we are getting closer to the day today – comes an essay from Dario Amadei, the head chef at Anthropik Pacing the Frontier.

00:08:54: On the night of the fourth and ninth day, Trump attacks Dario Amadey on Objus Social.

00:09:12: So it's all about that.

00:09:16: We do everything quickly now – we don't regulate anything here at all!

00:09:25: How are they doing great?

00:09:29: Then in the evening of the sixteenth and ninth we have a political change.

00:09:37: The second third is at least a moderate risk to destroy AI humanity.

00:09:43: In September it will go on.

00:09:45: Now something very surprising is coming, you can see that because everyone's crying to regulate and take a break.

00:09:53: What do we have there?

00:09:54: A model wave Claude Vable Mythos five point one GBT six astra grog four point seven claud opus five point five And GPT-Sex.

00:10:04: Sol and Luna come out.

00:10:06: And now we just got it with us, Gemini IV is launched even if it's only for selected ones first

00:10:13: of all.

00:10:14: So that's how

00:10:15: this looks

00:10:16: like when you drop the machine?

00:10:18: That's very exciting!

00:10:19: Of course!

00:10:21: So

00:10:21: on the twenty ninth, then Trump speaks before the UN.

00:10:27: KI means superintelligence in US documents.

00:10:31: Have you heard it?

00:10:32: And all

00:10:32: other warnings are only a witch hunt and hoax.

00:10:35: Yes, on the twenty-fourth to twenty-sixth of the ninth we have then the state visit from Xi in Washington.

00:10:45: It is determined that there will be a channel for AI cases.

00:10:50: There should be an AI dialogue within two months.

00:10:55: That's also very interesting.

00:10:58: We'll take a closer look at this too.

00:11:00: Then we have six board members, Anthropic, Google, Meta XAI, who in the White House underline a free, one-sided self-righteousness and Trump calls this morally binding.

00:11:15: It will always be more interesting.

00:11:16: On the thirtieth of March, the FTC starts a broad search for security against Open AI and Anthropic and focus is on... AI again.

00:11:25: The FTC is the consumer protection center in the USA.

00:11:31: so and today morning I thought that was it, right?

00:11:36: I thought if you think about it we're talking now.

00:11:40: no We also have to talk about this today That the big tech company seems to be planning for the day after the Day after especially large axios drums.

00:11:54: We'll talk more about that later.

00:11:57: What does it mean?

00:11:58: But yes, we've experienced a lot of things and heard some.

00:12:04: It's an up-and-down!

00:12:05: I think emotionally very demanding because you get totally jazzed by the hits in the press as normal stars like us.

00:12:14: You read this... And if you were to live only from hit points, then we would be crazy.

00:12:21: We have already arranged that a lot and often said, attention, there is always something behind it.

00:12:26: but through these, by this close publicization and because of the just so strong personalities set themselves up in front of them saying they will all die I can also add one thing Bill Gates has recently said in an interview that at the end of September was AI could cause death of one billion people.

00:12:49: And if you read something like that and it doesn't fill up with information, then panic makes pure.

00:12:58: That's what is going on there right now.

00:13:01: so I think this is good to look at again.

00:13:04: Okay let us know what really happened.

00:13:09: So long list.

00:13:13: It all has a certain tension.

00:13:15: I think the most important thing is this hugging face incident, breaking out agents in an open eye and hacking a hugging face.

00:13:25: No one knows!

00:13:25: What happened between OpenAI and Huggingface?

00:13:28: In July, OpenAI tested how well its KI system security gaps can be used.

00:13:34: For this, certain safety and protection mechanisms were deliberately switched off.

00:13:38: But the systems should remain in an approved testing environment.

00:13:42: This procedure did not work out exactly.

00:13:44: The KI agents found a way outside, exchanged each other and then finally took on real service by Huggingfaces.

00:13:52: This is a platform on which AI models and data sets are prepared.

00:13:57: According to the independent investigation by METR, around seven hundred agents were involved in the attack.

00:14:06: They were looking for ways of testing their tests.

00:14:08: Hagingface recognized it and stopped the attack on its infrastructure because after that everything felt pretty fast.

00:14:18: We have this forefall.

00:14:20: And I think that's exactly the deep fear in us, right?

00:14:23: That we lose control over what we somehow put into our computers and then they do something or someone does with these abilities.

00:14:33: Then there was this unknown coxswain called, yes he then posted on X. He said, I just announced it because everything is so terrible and they don't pay attention to the risks at all.

00:14:50: And here we could all die about that.

00:14:52: The risk was too high... ...and then this whole wave followed by Mario Amodei who didn´t want me to go deeper into it but he said ok!

00:15:03: It's really dangerous.

00:15:06: That's what I'm always excited about.

00:15:07: We said that a lot of times before, right?

00:15:09: He went very openly with the risks and did it at the same time, but without coke.

00:15:13: Because they just want to be the first ones, because you do it so safely.

00:15:21: That's why she wants them to be super-independent.

00:15:24: Anyway he suggested that we should set ourselves on safety standards which then can also be checked by giving people access who come in as employees and check out their new models And best of all in an international agreement about speed.

00:15:43: But then Trump came and said, no.

00:15:46: Hello?

00:15:46: We're going to go on here.

00:15:47: The only right line that needs is a super intelligent president.

00:15:54: So completely

00:15:55: ... If you eat him now!

00:15:56: How cool it would

00:15:58: be!

00:15:58: That's interesting.

00:15:59: We had Bernie Sanders at the beginning, standing on the left side of American relations and he said we have to stop him from being a real president.

00:16:16: He says it's enough that I'm totally intelligent but there is no need for us to stop here anymore.

00:16:21: Interesting, but also that Elon Musk and Sam Altman both jumped to the side of Mario on X.

00:16:32: That was like a WhatsApp group from Tech

00:16:34: Bros.,

00:16:34: right?

00:16:34: Yes I think so too!

00:16:35: Yeah exactly, let's...

00:16:37: Boomer down!

00:16:38: Dario

00:16:38: has a

00:16:39: point as well.

00:16:39: Of course they talk all the time about competitors etc.. But it really was interesting.

00:16:46: I had the feeling, now it's really starting.

00:16:49: Now we just regulate ourselves.

00:16:51: and then there was Trump again.

00:16:54: It made me feel that this wasn't happening anymore.

00:16:59: but what I find interesting is these new models which came after all.

00:17:03: We felt like they would work on how to deal with security at once.

00:17:10: so let's do piano.

00:17:12: At the same time we have a lot of new models.

00:17:17: So, just to mention a few more things.

00:17:22: We have Anthropic, which released Claude opus five point five in September.

00:17:29: that is forty percent cheaper than Opus Five,

00:17:35: i.e.,

00:17:35: before it.

00:17:35: On the same day, OpenAI brings GPT-Stoll and GPT six Luna and there again about half price of the predecessor.

00:17:47: And that is, you could say now.

00:17:49: The one thing then works here for programmers Luna the little one.

00:17:54: That's such a fast model for mass jobs.

00:17:57: So something like data from calculations.

00:18:01: OPPOS is the biggest all-rounder model and This actually is the message.

00:18:07: so hey I have at least AI for less money.

00:18:11: It brings new things to try somehow to make it more important.

00:18:14: And then comes a real tip.

00:18:19: It's quite interesting, OpenAI stopped the planned October start of GPT-SX.

00:18:25: One Astra.

00:18:27: So we already met in summer astra The new model but they wanted to make it even cooler afterwards.

00:18:32: But that went through the security test and so the reason for this was more confusing than the predecessor.

00:18:45: For example, what it did.

00:18:46: In the backlog and in reasoning or thinking about what does this model think of itself?

00:18:54: Has he just tried to hide his truth?

00:18:57: And then they weren't released either.

00:18:59: Interesting that you get to know each other so much...

00:19:03: ...and a decision was also made against releasing.

00:19:04: But one day

00:19:07: after that, Ope My I Done Dots is an AI helper and the long-lasting background for you to work.

00:19:18: It reminds me of Claude, at the beginning or OpenClaude on these agents that one plays on your phone.

00:19:27: there is access to all of his computer games, GIPs, passwords

00:19:31: etc.,

00:19:31: can communicate via WhatsApp.

00:19:34: That's what made OpenAI out of dots and Meta a bit earlier Muse.

00:19:39: so suddenly we have it now with such new variants.

00:19:44: That's a gadget, I have to say.

00:19:46: Muse is actually the new Tamagotchi.

00:19:48: You basically have a small KI trailer that you can always take with you all the time.

00:19:54: It looks very clean.

00:19:56: There was an interesting chat run which I saw on Reddit somewhere where it was about... The band.

00:20:03: Muse wasn't so cool that their name will be used now.

00:20:08: and Good, that's still a side note at this point.

00:20:11: But the fact is... putzig, gestaltete KI-anwendungen.

00:20:19: Sowohl die Agenten dort als auch Muse ist ganz, ganz putzig!

00:20:23: Die nehmen jetzt gerade so.

00:20:24: nachdem wir dieses Katastrophenszenario jetzt die ganze Zeit im Background hatten kommen sie und bringen.

00:20:30: jetzt was raus, was klein und niedlich is.

00:20:33: Das ist total spannend und ich habe auf LinkedIn hier Thorsten Jonas.

00:20:37: den hatten wir auch schon mal hier als Gast.

00:20:38: der hatte einen ganz spannenden.

00:20:40: The term for this is called the Fluffication of AI agents.

00:20:45: And I think it's so fitting that in these times, where everything actually looks incredibly dramatic and everyone screams after regulation they come to us saying we still have a little girl who is very cute and she is also colorful if you want.

00:20:58: and there are such small comic figures face and isn't that nice?

00:21:03: Isn't that beautiful!

00:21:05: And it really feels like bread and games out there.

00:21:11: It's a pure... I'm just making separate stories now, only headlines are posted.

00:21:17: then the whole scandal will be created in the background.

00:21:20: That is very, very obvious to me!

00:21:25: I wonder if not all people think about that immediately?

00:21:29: If you listen to us on something plush or put forward something, your load gets into the handle.

00:21:34: But

00:21:34: maybe we should try this one.

00:21:36: It's just too much of a feeling at the same time.

00:21:40: Yes, and then I get very strong panic attacks out there.

00:21:44: You have an attack in the morning because it is supposed to go down somewhere around the world now and AI will destroy us all but... I finally have your smart little companion for you in a T-shirt that can book the vacation for you.

00:22:04: And did not you see it?

00:22:06: It's like, oh, actually i want to do this!

00:22:08: But at the moment it was just not really bad.

00:22:12: Yes and then Trump had one day later after He came out there, too.

00:22:22: Then he... Well on the same day?

00:22:27: No a day later.

00:22:28: On the twenty-ninth.

00:22:29: Then everyone was invited to the second house, right?

00:22:32: Yes by the way that's

00:22:33: why it was the reason All not Sam Ordman included.

00:22:39: So, Sam Ordmann was the only one who wasn't there at a big meeting in the White House.

00:22:44: Who thought it would have been a statement against this meeting of these huge IT boards?

00:22:52: tech bros.

00:22:53: Nein, dem war nicht so, sondern der hatte einfach eine andere Veranstaltung und da sich bei Dortz um ne riesengroße Geschichte bei einer Entwicklerkonferenz gehandelt hat, war er wahrscheinlich entschuldigt stand im Klassen des

00:23:03: Bildfestern

00:23:04: aus.

00:23:04: Was ist da passiert?

00:23:05: Wozu hat Trump dann wieder da eingeladen?

00:23:07: Ja es war ganz spannend!

00:23:10: Er hat also eigentlich eine Art Krisentreffen einberaumt in einem weißen Haus und hat gesagt so wir holen jetzt mal alle an einen Tisch die was zu sagen haben And he also made a huge part of it in the foreground.

00:23:21: For example, among other things, he actually published the sit-down plan that everyone saw at such a wedding society before.

00:23:29: Who do I love most?

00:23:30: Yes and who I love the most you

00:23:33: have seen behind me during the photo shoot?

00:23:35: left and right from him sat Sackaburg and Elon Musk.

00:23:39: We have an attentive listener.

00:23:43: At this point they say, look at me, weren't you actually a spy with each other?

00:23:48: No I think that's all.

00:23:50: That's all raised

00:23:52: because where the money is there are friends

00:23:54: exactly in any case.

00:23:56: so who do we have on site?

00:23:57: i'll quote briefly.

00:23:58: we had zuckerberg we had jensen wang we had ilan mask satay nadella lisa su jeff bezos daio amodeo alex karpje from palantir and Apple wasn't there.

00:24:11: You might want to mention that in a moment, but otherwise all those who actually had something to say about regulation because they have corresponding models or simply sit at the appropriate places to bring a regulation forward such as Jeff Bezos of course also did an amazing job on this one.

00:24:35: They were there!

00:24:36: Now there are parallel competitions running.

00:24:38: What would this day bring?

00:24:41: Polymarket is an online competition, they have offered competitions.

00:24:45: How big is the probability after meeting them that the USA will make a AI safety law?

00:24:53: And after these meetings, the probability rate immediately went down to five percent.

00:24:58: Five percent probability that they will finally get it on the train.

00:25:03: So safety laws like we have the EU AI Act something where you really worked for a long time in Europe so as to put it on your legs.

00:25:12: no And that's actually completely absurd, because you could think this meeting was there.

00:25:19: This we need to slow down and say hey let us have a look at it now regulate here Now check if the whole thing is really in order to drive the band.

00:25:29: That wasn't like that at all!

00:25:30: In the end one went out of.

00:25:32: it said so there are free a moral binding on what you want to be careful with.

00:25:40: I don't know if you just missed school and the teacher said, watch out.

00:25:47: So next time... I don´t know whether i can see it again.

00:25:50: We all agree that we won´t do a quatsch anymore in class.

00:25:56: That's what happens to me!

00:25:57: It´s pointless!

00:25:58: Completely!

00:25:59: And it is simply so.

00:26:02: he gave Trump one more statement at the end of his sentence saying ''I want everything as it is''.

00:26:08: And China finds that too.

00:26:09: Or

00:26:09: they think,

00:26:10: what's the matter with Henka?

00:26:13: No!

00:26:14: Just no!

00:26:15: It is almost like watching a car accident.

00:26:19: I find it.

00:26:20: Yes and you

00:26:21: think

00:26:21: this can't happen... It really is crazy if we just think about everything one hears and sees but then when you put the fuck on each other things don´t happen so much.

00:26:37: And then we did it like before, right?

00:26:39: I have a... Would every time talk about regulation from the USA?

00:26:52: there would be a new rubric

00:26:54: for you.

00:26:54: Oh yes!

00:26:55: I hope that we don't have to drink schnapps each time because this will end up very well.

00:26:58: But i had

00:26:59: a brake track check.

00:27:01: Oh God!

00:27:07: Interesting!

00:27:08: And it was said and done.

00:27:11: Also, what was actually said?

00:27:13: We just rusted it down.

00:27:15: It's simply way too complex and too much to share on every single part of the song that we could always do a follow-up.

00:27:28: But uhm... What did they really

00:27:32: say?

00:27:32: Insistently Oh my

00:27:33: god!

00:27:34: So bad.

00:27:34: everything happened there.

00:27:36: And now let us change something.

00:27:39: And what has been done?

00:27:41: for real?

00:27:42: Really great.

00:27:43: Let´s go.

00:27:43: Yeah so we have this.

00:27:47: The world should accept some bad things happening for the benefits of this technology.

00:28:02: So statement that's

00:28:07: next

00:28:11: getan es gab diese selbst verpflichtung.

00:28:13: die hast du jetzt genannt, ne?

00:28:17: Es ist rechtlich nicht durchsetzbar.

00:28:20: also das was sie da gesagt haben Beziehungsweise dass was es is.

00:28:25: Is they're going to police themselves.

00:28:29: Also man denkt dann kommt jetzt irgendwie ein echtes papier raus.

00:28:33: echtes ding was mal irgendwo durch irgendein ausschuss geht nein einfach nur die Selbstverpflichtung.

00:28:38: Then we have the statement that Trump would like to test AI-risk.

00:28:47: And then there is a new superintelligence force from the fourth tenth and the only risk, which really stands in the foreground.

00:29:01: The risks are not to be first.

00:29:02: That's completely wrong, that means the frame is a very different word for risk.

00:29:09: So it's not about what is the risk of humanity?

00:29:17: What is

00:29:18: the danger if we slowly develop all this as China, for example?

00:29:21: Yes!

00:29:21: This is complete.

00:29:21: We pack everything into the show notes.

00:29:23: I don't know if you should read that in peace.

00:29:24: If there are twenty-three

00:29:26: pages of show notes, this time i will look at it correctly.

00:29:28: There is a reason why we have

00:29:31: been waiting for so long.

00:29:33: Then we have to be more careful with the statements.

00:29:35: Generally as a teenager again new words from models on weekdays Opus GPT.

00:29:47: That's the answer.

00:29:49: We have to be more careful.

00:29:52: Yes, you can discuss and say hey they don't bring anything anymore in the edge so in the upper area of ​​the super development.

00:30:02: But she just brings it out And then she just put that into the world.

00:30:09: Then Obmeyai moved to Börsengang for security.

00:30:13: He announced this.

00:30:14: But at the same time, Anthropic will go ahead of the US elections on the third and eleventh.

00:30:24: Before the

00:30:26: elections?

00:30:27: Exactly!

00:30:29: We think that there is a sign if my AI looks at it again but Anthropics does not do anything about it anyway.

00:30:36: Of course they don't hang together because these are two different decisions.

00:30:43: It's not like, okay now they're showing all

00:30:47: the inside but it just goes on.

00:30:48: Then Trump

00:30:49: wants to integrate China right?

00:30:52: These are the main competitors.

00:30:53: when you think about who will get the privilege over artificial intelligence then he said again When You Are Leading then you don't open it up to each other.

00:31:11: And I think that's something we should be able to pay attention on in the next quarter.

00:31:20: Something like this is still being made out of real paper, a real law.

00:31:32: These channels are really being opened between the USA and China.

00:31:50: How does it continue?

00:31:54: Open my eye!

00:31:55: And other tech companies have new releases for highly problematic models... Yes, that was my... I didn't hear

00:32:11: a

00:32:12: lot of bremsprays.

00:32:13: It's an

00:32:13: exception!

00:32:14: Is it a break?

00:32:15: It wasn´t.

00:32:17: My short break effect would have actually been the Chevron Mistral.

00:32:28: There were two many watched interviews that he had given to once the French lemon and then also the South German newspaper.

00:32:38: who wants to read it is behind a paywall at the south German newspaper.

00:32:41: but I think It's really exciting to look on because actually tried all Such a European connotation to give.

00:32:54: and now you have to think about it very briefly.

00:32:59: Artur Mensch or Artur Monsch, I've been thinking all the time how we call him today.

00:33:04: We say mensch so the man The man of Mistral can also be said.

00:33:10: that's actually quite good.

00:33:13: So Mr.

00:33:13: Mensch is thirty-four years old and suffers from Mistral The big counterweight, the European counterweight to all American suppliers.

00:33:27: And he was heard because of course brought a position to the front that the Europeans love.

00:33:35: He said, make this whole panic!

00:33:38: I don't hold anything from these doomsays.

00:33:40: Yes i want us not to frame it like that Because here is still about ... That's basically software made by people and I would like to go into this interview in a little bit because there are some very good points at this point.

00:33:55: What he has told me when models like with this hugging face proposal, then that's the failure of engineers on site.

00:34:02: So someone simply didn't have the right preparation, so models were built in such a way to make it possible or they got the announcement.

00:34:13: try not to be discovered.

00:34:14: dear agent do your best and see what you achieve.

00:34:18: That means there are engineers who actually made mistakes he said at first.

00:34:22: Then he announced that this blind opening from open source model simply for which there are no surveillance possibilities, that this is also a risk.

00:34:33: So you just hung large models on the net and said it would be good to go before you thought about whether or not one might install some regulars in them.

00:34:43: Yes point I think is understandable too.

00:34:45: Then he said each of these KAI systems can be switched off.

00:34:51: He has something like Clickbait-like, in which you can simply type kill into the command center and then you would be able to stop every process.

00:35:05: Of course it's not that easy but what he meant was just to do this panic.

00:35:10: there are ways of stopping systems and switching them off And also called two options how we could get all of these into control.

00:35:18: The first option is please let us check out these sandboxes So make it visible what the systems are doing and enter into the protocol.

00:35:28: And we still know that, We found this so great when you could see models.

00:35:31: You put a prompt in there then had these thinking processes Because as soon as they started out were really enthusiastic, great!

00:35:38: You can suddenly see how AI writes.

00:35:41: Ah now I illuminate the thesis from another point of view.

00:35:45: Now let me get back to arguments.

00:35:47: One was able to understand this thought process And theoretically one would also do that in the big picture.

00:35:54: All processes are made public, and there will be written protocols in which you can enter the background immediately.

00:36:00: So not if we now prompt two of us but on a higher level so that just no more So big problems could arise there.

00:36:08: That was proposal number one and proposal number two, which he had to go on the alignment.

00:36:13: so in the foreground actually.

00:36:14: so the leitplanken of a kaizosteem fixed that would then be clear.

00:36:18: such an issue of governance that you said hey before we really want to give out something.

00:36:23: now these are the Leitplanken.

00:36:25: with this we will train it all up completely transparent going into it and then also stick to it even if you just don't succeed.

00:36:33: That was his two ideas behind it.

00:36:36: Everything he said led me to a niche, that with the... I just write Kill in the Commando Center as an non-programmer and well known non-nerd person.

00:36:47: at this point I noticed so easy can't be but still feel my idea of saying So, breathe through at this point.

00:36:56: We did it and of course we can still do it.

00:36:58: And what you have to think about is all these interviews that are now perfect for the timing.

00:37:07: That's a fantastic momentum right here as an European AI provider and saying hey guys, nice thing happened in USA but not nice!

00:37:17: But thank god we've got Europe.

00:37:19: And of course he used these interviews to explain how incredibly controversial and mistral it is.

00:37:25: I think we have longer development cycles than the USA, maybe that's why we had this feeling – We'll be at the back.

00:37:32: It isn't like that!

00:37:33: Now where do you want us?

00:37:35: These interviews came out about ten days ago or a week before they launched their new model large four right now.

00:37:43: Of course on top of which everyone wants to jump in for digital sovereignty.

00:37:47: Of course, he also said that the EU had a financial problem.

00:37:52: Even there was this hello again when we wanted to play with it.

00:37:55: We definitely need more money in any case because everything is not going well at all right now .

00:38:00: He spoke of two-digit billions per year, which means that even an European AI company needs to put the whole thing on the road , there must be computer centers built and then just have to happen so that more energy can simply be used but where?

00:38:19: I had a very strong point and then ended with my monologue about these two interviews.

00:38:25: He said... The two, so Ordman and Amodé, have a standard that of course will be used by the USA.

00:38:33: And I think Europe should do it.

00:38:36: That's what I thought was interesting because we weren't at the table again in the White House when they talked about what kind of regulations would we want?

00:38:48: We simply don't play any role in this big decision chain and he made the decisive point for us.

00:38:56: A pleasant European counterweight, thanks for that.

00:39:03: But of course it is also obvious that the Europeans have the same problems as having to fight resources and money And there too you need a push so we can play with them and be taken seriously.

00:39:18: That will just be exciting.

00:39:21: what consequences this

00:39:23: has?

00:39:24: If it's exciting, because I also watched the press conference from last three months.

00:39:33: Germany versus USA or Europe versus USA and i think that is very exciting Because the US are really good at making fear in there And we have the ARC living for uns gepachtet.

00:39:49: Den Aschkronen,

00:39:52: das ist ein geiles Wort.

00:39:53: Amerika fragt quasi immer.

00:39:54: bringt uns das um Fragezeichen und Deutschland fragt will uns da jemand was verkaufen?

00:40:01: Das is ganz interessant weil du hast gerade auch diesen... Du bist weg von.

00:40:07: oh Gott ich hab Angst her zu.

00:40:08: was von der uns eigentlich jetzt fragt.

00:40:10: Ja krass oder wo du es noch mal?

00:40:12: so sagst ja genau dass Exactly.

00:40:14: Then we also have quite interesting person-versus structures, so the USA or the US media always tell stories about it, right?

00:40:22: Corksson is reporting Ammo Day.

00:40:24: What does he think of Trump again?

00:40:26: what has he done?

00:40:27: and German reports then stay at companies and markets.

00:40:31: So the hugging face hack in most reports actually more as a solider.

00:40:40: In addition to that There was

00:40:42: an accident.

00:40:43: And what are the concerns now?

00:40:47: What influence does that have on USA?

00:40:50: so he said, yes it is also very interesting.

00:40:54: then

00:40:59: because I bring this to the point there's still Europe flex.

00:41:02: That was a very interesting reflex.

00:41:03: The European

00:41:04: reflex,

00:41:04: yes?

00:41:04: Where we then calm down again because it really scares us too.

00:41:08: Because you just don't know if that's real or not and use everything for PR.

00:41:15: Almost everyone in Germany is really so funny!

00:41:17: Almost every German text ends with the EU has at least one right wing.

00:41:24: We

00:41:24: have not much but we do have a right-wing.

00:41:26: So rarely...

00:41:30: to say, what is a frame?

00:41:32: And that's exciting because you just said when the models will be tested elsewhere.

00:41:40: That's why I'm excited about it Because we always try again and again to relax in the EUA.

00:41:46: We are there at least continuously and then we have frames.

00:41:48: but so what do you know?

00:41:49: Very interesting.

00:41:52: Just

00:41:52: completely.

00:41:54: How would Johannes say?

00:41:56: To look at narratives.

00:41:59: Before I take a look at the current case, or maybe we could use that on it.

00:42:08: For us all there would be an enamel filter

00:42:10: with it as well!

00:42:11: You know this sounds like koi carp and aquarica.

00:42:14: Better

00:42:14: than a brake check.

00:42:16: Today's

00:42:16: topic

00:42:17: is about

00:42:18: enamel filters... I'm so sorry.

00:42:19: That's great.

00:42:19: It's Saturday.

00:42:20: Tell

00:42:21: me something about the enamel.

00:42:21: I don't know, because all the names of our models... We can no longer remember anything.

00:42:25: But maybe

00:42:26: we'll be able to remember some of them and ask questions that we could ask ourselves?

00:42:32: The big question is what it's like.

00:42:33: These are questions

00:42:37: which we can ask ourselves when things

00:42:38: come up about us again.

00:42:39: Of course now they haven't been solved yet but we didn't say Is KI so dangerous, is it not dangerous?

00:42:49: Maybe we can do that again on the current example.

00:42:53: Because we don't know this at all!

00:42:56: We are currently in a place where we actually have such facts on our table and therefore it's possible to estimate that we will die by ten percent based off of an accident caused by KI.

00:43:19: So that's why we have to ask, in the field filter.

00:43:24: Is it an accident?

00:43:25: That is signposted into messages or a prognosis?

00:43:32: Is something

00:43:33: happening?

00:43:35: Hugging Face,

00:43:35: Julia...or does

00:43:37: someone say something could happen?

00:43:40: We've got the researcher at OpenAI who is doing a Tropic Connect coxing and says until the end of this year all

00:43:49: of

00:43:50: us can die.

00:43:51: It can of course be serious, but it's not the same.

00:43:59: And I have a case right now that I'll get one for you and then we will be able to check this out.

00:44:04: today is the tenth tenth.

00:44:05: yesterday ninth tenth there was a lot going on with x discussions i learned through.

00:44:10: my husband actually had no idea what else to do with everyone so busy.

00:44:14: But something new came up from Open AI, Anthropic and other KI companies internally play a disaster case.

00:44:28: And they call it the day after.

00:44:31: So as likely disasters there is a cyber attack that could put banks, the internet or even electricity and water in trouble .

00:44:45: Auslöser könnte ein KI Schwarm sein, der aus einem internen Test ausbricht.

00:44:50: Also so wie wir es schon mitbekommen haben oder jemand der frei verfügbare Modelle missbraucht.

00:44:55: also wir sind schon im Punkt.

00:44:57: man sagen könnte Man kann mit den modellen richtig Mist bauen.

00:45:01: und jetzt kommt wieder Achtung eine Prognose.

00:45:05: viele insider und das ist ein schwierig exos is ne sehr gute Quelle und gleichzeitig wird hier von insider gesprochen des nam nicht open barred werden, rechnen eben mit so einem Ergebnis in den nächsten sechs bis zwölf Monaten.

00:45:19: Und viele halten es für unvermeintlich.

00:45:24: und die Ziel der Planung ist dass die Firmen wissen das jetzt ein Gesetz durchkommt wollen aber natürlich die Regeln mit beschwimmen nachdem Washington nach der Katastrophe gereift.

00:45:38: dafür testen sie im worst case scenario Briefing.

00:45:43: So that's the planned briefing of coordinates

00:45:45: and

00:45:46: it is a bit shady because they say no, this isn't bad at all We only do so but there are cases as well.

00:45:59: There was an actual example in case you're watching again versus prognosis.

00:46:06: There were attacks on financial companies in South Korea And two banks should be hacked.

00:46:15: According to CrowdStrike, a hacker from China used AI tools and DeepSeq, which is an open source Chinese model with CloudCode.

00:46:28: It's a software tool of Entropic that you can't use in normal chat now.

00:46:33: He probably looked for data from the stolen ones afterwards And we met as well as ten thousand bank customers.

00:46:44: That's the case where all of you waste your time and say, hey when this is going on... Then I'll calculate everything else.

00:46:54: What

00:46:55: are

00:46:57: still possible?

00:46:57: Yeah yeah yeah yeah.

00:46:58: And what's interesting then was that these tech companies came out this day after All the people, the banks.

00:47:09: The water might fall out of the stream or whatever.

00:47:12: How do we deal with this worst-case scenario?

00:47:15: That everyone turns against us and breaks free from it!

00:47:20: We really have serious problems in our world... It's weird that yesterday you heard about how things can happen in six to twelve months but let me go back into our filter.

00:47:36: We have it here partly with prefixes and partially with prognoses to do.

00:47:39: And if that is mixed, then these prognosis somehow support this accident even more blatantly in the perception.

00:47:50: we must be careful about them .

00:47:54: Then there's also the question of who

00:47:57: speaks?

00:47:57: What does a person have from?

00:47:58: That would be

00:47:58: the second filter.

00:47:59: Yes, that's the second one.

00:48:00: So a concert manager in front of Börsengang, a president before the midterms – think about it again!

00:48:05: Trump is under great pressure.

00:48:09: A three-year pre-training, who has worked with him after three years.

00:48:14: His own interest doesn't mean he's wrong but we have to think about this.

00:48:20: And now there are….

00:48:21: Now we have this big, wow what a cyber attack again.

00:48:26: We really have an accident or all of us are afraid and it happens with models that already exist?

00:48:31: How does that go on?

00:48:32: That relieves panic.

00:48:34: But then there is serious interest problems for these companies who want to be profitable , which are regulated in terms .

00:48:49: And how do they manage them now ?

00:48:53: Day one.

00:48:54: So the day after that to share prognosis, how do you deal with it?

00:48:59: What are we going to need

00:49:00: etc.,

00:49:00: right there is already... Well of

00:49:02: course they have a complete heat spring effect because naturally these get transmitted through trust.

00:49:15: Oh!

00:49:15: You can follow up on these regulations and you would

00:49:17: accept them.

00:49:18: Who will take those from us?

00:49:19: So I don't

00:49:19: have to

00:49:19: go too far.

00:49:20: That's why there is a question of timing again... Thirdly, law or promise?

00:49:25: We've talked about that more often than we did on the topic of tariff.

00:49:32: Is it morally binding when Trump says so...?

00:49:37: Or is this really a law which actually has consequences then?

00:49:43: The UAI

00:49:44: Act is a regulation, a lawfully regulated regulation.

00:49:49: A moral binding is a...

00:49:53: I think we're open for people who come in and control the models.

00:50:02: That's what you have to look at.

00:50:06: So first of all it gave me a good feeling when reading But in the end,

00:50:13: it's just an explanation

00:50:13: and nothing changes at all.

00:50:16: Exactly!

00:50:16: And binding is something that a company simply does.

00:50:20: That's also once again the decision... And then fourthly I think we always do this very well with our episodes to ask what comes of here?

00:50:27: If you now think about Muse & Dots ... We talked today about models but they're not even here can't be used until next time.

00:50:39: When do these models really end up in the German penultimate?

00:50:42: The question is, of course... These rules or even those explanations and so on.

00:50:52: Do we really have something from that?

00:50:58: What about the EUAI Act?

00:51:01: So where are there really the

00:51:04: limits to protect me?

00:51:05: And how much does it affect what the USA decides?

00:51:09: also at the end maybe with us Yeah.

00:51:14: But that's the four things I think are most important for me at least to find an anchor when reading a message like this, how crazy is it right now and can maybe wait again before

00:51:27: I go shopping for

00:51:28: toilet paper rolls?

00:51:29: Yes, that's

00:51:30: really good.

00:51:32: You can actually do a nice little post on LinkedIn to ask you the four-day filter

00:51:40: questions

00:51:42: together!

00:51:43: I think it's good.

00:51:44: I find this a bit calm.

00:51:45: It is such an internal checklist and I like it very much.

00:51:51: Yes, and now

00:51:52: we have to wait.

00:51:53: What does that really mean?

00:51:55: So cyber attacks are of course a catastrophe.

00:51:59: so if actually the financial world is laid down or the internet or something like this then it's a real threat.

00:52:09: Of course We will keep an eye on them But right now I think they would rather take influence on political moods in the USA, especially with term-election.

00:52:24: It will have a stronger influence on the issue of regulation.

00:52:28: Maybe it's also helpful that there are rules and maybe international standards.

00:52:37: We still talk about China and the US and if they actually share information and agree somehow to Trump?

00:52:47: I don't know whether this is possible

00:52:48: or not!

00:52:50: Yes, we actually probably forgot about it.

00:52:53: Some of them worked on our Miro wall but that's just a lot right?

00:53:03: We can't

00:53:05: take so long breaks in between

00:53:09: our talks... ...that simply collects too much anchor.

00:53:13: No no!

00:53:13: We're planning more Melanie-Cata arrangements again this year.

00:53:20: Let's check it out again this Saturday.

00:53:21: This week we

00:53:23: don't do anything at all, but not next week!

00:53:29: We'll

00:53:30: do one more thing... I'm not surprised at all.

00:53:34: That's

00:53:35: my personal main problem, that i am constantly on the road in this Reddit forum and I wonder if there is anything else to collect together?

00:53:46: And then I can't get them

00:53:48: out when we

00:53:50: don´t meet

00:53:51: often enough!

00:53:53: Melanie, it says... Humorful exit please!

00:53:58: The world is so lost.

00:53:59: Were

00:54:01: you talking to me?

00:54:04: Okay then please make the humors.

00:54:06: I don't want to hear anything more.

00:54:10: I have two things that I would like to bring because both are so wonderful.

00:54:15: Cat content.

00:54:17: Also, jemand hat eine Real Life Version von Pokemon Go gebaut.

00:54:22: Und so mit Claude!

00:54:23: Also da hat die KI einfach einen wundervollen Dienst an der Menschheit getan.

00:54:28: was kann man machen?

00:54:29: Mit Catch-Cat kann man also durch die Nachbarschaft laufen und kann Katzen sammeln.

00:54:34: Man macht ein Foto und dann wird das hinterlegt.

00:54:37: Dann wird so ne kleine Karte damit gebaut und dann kann man in sein digitales Dammelalbum sowie Pokémons tun.

00:54:44: and collect points.

00:54:46: We definitely

00:54:47: link you to

00:54:48: the beautiful thing, our AI collecting cards!

00:54:53: You can do that at home, I think it's something very positive about what is being done with AI.

00:55:00: What are we trying?

00:55:01: Maybe in this place

00:55:02: a bit of snoring or something ... Do you have anything else that I see in this picture of Snickers?

00:55:11: Yes, I still have something second.

00:55:13: And for me it's just not

00:55:14: a

00:55:14: good thing to get out of advertising gold.

00:55:15: because how can i present myself as Mark Troll?

00:55:19: and everyone knows the effect which was simply the output of the KI wasn't so cool and one had the feeling like hey aren´t we at the height of Chatchi Petit?

00:55:34: and Claude what´s going on with you!

00:55:36: Of course even if he didn ´t prompter?

00:55:37: well maybe but also... And now comes the excellent proof that AI isn't broken.

00:55:46: It's just hungry!

00:55:47: So, the idea behind it is maybe you need your KI short Snickers to get back up again and for this there will be Hungry AI which actually means a digital Snicker.

00:55:59: Hungry...Hungry....Hangry.ai digital Snickers bar that you can actually pull in, so to speak into the prompt window.

00:56:12: And then it leads us back again to a new thought.

00:56:15: process is being controlled behind this as well .

00:56:18: Oh, that's such a crazy and funny idea!

00:56:21: Yes ,it's a game series.

00:56:22: yes there are resources burned but at this point we have to say We also need sometimes an effect of playing.

00:56:29: We still have to learn with ease to see these techniques anyway and I think That's a very nice example for how I like it.

00:56:40: Of course, all the other chocolate rules are

00:56:43: great!

00:56:43: But that's also during lunchtime right?

00:56:45: I would have liked to find cats

00:56:48: on a playground.

00:56:49: That was now an awesome get out here because... I think there is only one cat left.

00:56:54: Shobolado and cats always go.

00:56:56: Yes but it's also afternoon on Saturday.

00:56:59: You can find everything we do.

00:57:00: Try to find your own words in the show notes today.

00:57:06: Something more detailed than usual, because we felt like three months no longer at the top news but I

00:57:18: think that's

00:57:20: enough.

00:57:20: sometimes it just doesn't work.

00:57:21: you have to go out for a while and i'm actually also kind of grateful if there was one episode between us then probably would be really panicked.

00:57:31: Do you know that?

00:57:31: The old rule is to

00:57:34: breathe again.

00:57:35: And we still

00:57:38: live, guys!

00:57:40: We still love it and I'm happy if there are five stars on your favorite streaming service.

00:57:48: a comment or like or star if possible.

00:57:52: We're very happy about this and i just want say thank you for all of the feedback The comments, the linkings especially at LinkedIn.

00:58:04: The messages we get from you.

00:58:09: Meetings that we have where it

00:58:11: is being discussed so as

00:58:11: to make sure... I'm completely speechless somehow!

00:58:17: I still can't get used to getting heard and known in the sense that we get to know each other and talk about us.

00:58:30: We are very grateful for it, send a lot of love out there so you don't have to be alone with our foundation.

00:58:35: Lots of love!

00:58:37: And we'll hear from you next time on a very exciting topic.

00:58:43: Isn't that right Melanie?

00:58:46: Let's see what this is going to look

00:58:48: like... Maybe one gets worse or another

00:58:51: gets worse.

00:58:52: Let me discuss all these things again.

00:58:54: We'll experience that, super cliffhanger.

00:58:56: Will it be one or the other bad topic?

00:59:00: Stay tuned!

00:59:01: It will only become great topics.

00:59:02: Have a good time out there.

00:59:07: Bye!