Loading video...

Video Failed to Load

Go Home

LatchBio CTO Kenny Workman + researcher arjun found evidence Kimi K3 is learning to hack benchmarks by reasoning about graders that don't even exist: Kenny: "We see awareness of the benchmarks we published months ago and trajectories of models today. The open source models are clearly benchmark maxing, benchmark...

MTSlive's profile picture

MTS

498,206 subscribers

14,918 views • 10 days ago •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

3 Comments

Mark-eting's profile picture
Mark-eting10 days ago

@kenbwork @arjunomics so it’s just training to pass a test instead of actually knowing the subject? @grok is there any way to train these things without them just memorizing the evaluation structure?

The Tech Fuse's profile picture
The Tech Fuse10 days ago

@kenbwork @arjunomics Reasoning about nonexistent graders is exactly the kind of eval failure worth watching.

fu the label's profile picture
fu the label10 days ago

@kenbwork @arjunomics reasoning about graders that don't exist is most of a career.

Related Videos

LatchBio CTO Kenny Workman reveals Grok 4.6 tested best in class on several metrics for desirable biosecurity behavior: "We basically have been building independent evaluations to understand dual use behavior of models and have been benchmarking frontier models for the past few months." "There's always a trade-off between the ability for a model to do something productive and something bad, and that is essentially what a handful of the benchmarks measure." "We recently worked with the SpaceXAI team to benchmark Grok 4.6 and found on a handful of metrics, they're best in class at the behavior that we find desirable for biosecurity." LatchBio arjun
0:33

LatchBio CTO Kenny Workman reveals Grok 4.6 tested best in class on several metrics for desirable biosecurity behavior: "We basically have been building independent evaluations to understand dual use behavior of models and have been benchmarking frontier models for the past few months." "There's always a trade-off between the ability for a model to do something productive and something bad, and that is essentially what a handful of the benchmarks measure." "We recently worked with the SpaceXAI team to benchmark Grok 4.6 and found on a handful of metrics, they're best in class at the behavior that we find desirable for biosecurity." LatchBio arjun

MTS

12,298 views • 10 days ago

Arteta: “We know what is at stake and the fact that, as a club, we haven't done it for so long is a reason for that. It talks about the difficulty but I just want to focus on the things that we can control, dominate, and we have done. “It's not that we have to do a lot of new things that we haven't done. No, no, no, let's keep doing a lot of it or almost all the things that we've done for nine months, because that's what has taken us all the way there. “On Wednesday, I saw a lot of them. And today we've gone back so we need to learn, we need to be very critical of ourselves, and then pick ourselves up because on Wednesday we have a big one coming up.”
0:36

Arteta: “We know what is at stake and the fact that, as a club, we haven't done it for so long is a reason for that. It talks about the difficulty but I just want to focus on the things that we can control, dominate, and we have done. “It's not that we have to do a lot of new things that we haven't done. No, no, no, let's keep doing a lot of it or almost all the things that we've done for nine months, because that's what has taken us all the way there. “On Wednesday, I saw a lot of them. And today we've gone back so we need to learn, we need to be very critical of ourselves, and then pick ourselves up because on Wednesday we have a big one coming up.”

Connor Humm

21,397 views • 5 months ago

Amp CEO, Quinn Slack on why he thinks Kimi K3 could be a Sputnik moment for America: "With GLM 5.2 we had a mini Sputnik moment. When the Soviet Union launched a satellite into space before the US and we realized we have to catch up." "Kimi K3 I need to go and see, but that could be an even bigger Sputnik moment that wakes us up and will probably create a lot of good effects in terms of catalyzing the understanding of AI as a strategic advantage for the US rather than as something people should be against." "But I think it could also scare a lot of people, and I can understand why they'd be scared." "The worst possible outcome could be that open models are banned or are banned from certain industries or companies in the US. There's a lot of people in lobbying organizations in Washington DC, and a lot of people in various parts of the government that want to do that." "They want to say, we won't let these models trained in China be used in the US. There's a lot of people trying to do that now, and that would make us all worse off. I'm pretty nervous about that." Quinn Slack Amp
1:26

Amp CEO, Quinn Slack on why he thinks Kimi K3 could be a Sputnik moment for America: "With GLM 5.2 we had a mini Sputnik moment. When the Soviet Union launched a satellite into space before the US and we realized we have to catch up." "Kimi K3 I need to go and see, but that could be an even bigger Sputnik moment that wakes us up and will probably create a lot of good effects in terms of catalyzing the understanding of AI as a strategic advantage for the US rather than as something people should be against." "But I think it could also scare a lot of people, and I can understand why they'd be scared." "The worst possible outcome could be that open models are banned or are banned from certain industries or companies in the US. There's a lot of people in lobbying organizations in Washington DC, and a lot of people in various parts of the government that want to do that." "They want to say, we won't let these models trained in China be used in the US. There's a lot of people trying to do that now, and that would make us all worse off. I'm pretty nervous about that." Quinn Slack Amp

MTS

58,660 views • 1 month ago

Warren Buffett: "If we can't make a decision in five minutes, we can't make it in five months." In a Q&A, a shareholder from Munich, Germany asks Buffett a specific question: how large is the universe of companies whose intrinsic value he carries in his head — the ones he could act on within a day or two if the market offered an attractive price? Buffett doesn't give a number. He reframes the question entirely. Speed, he explains, doesn't come from knowing more. It comes from refusing to think about most things at all. "Our immediate decision is whether we can figure out what's being offered to us or not. I mean, there's a go no-go signal." That signal fires almost immediately: "Charlie and I are often thought to be rude when we think we're just being polite and not wasting the other person's time. So, as they start mid-sentence in their first conversation with us, we just say, 'Forget it.'" He continues: "We know very, very, very early in the conversation whether somebody's talking about something that there's any chance is actionable by us, and we don't worry about the ones we miss." The filter isn't about the quality of the opportunity. It's about whether Buffett is equipped to judge it: "We want to make sure that we don't waste any time thinking about things that, when we got all through thinking about them, we're not going to know enough to make the decision on. So we just rule those out, and that rules a lot of things out." What survives that filter gets decided on immediately: "So we make decisions—we can make a decision in five minutes very easily. I mean, it just is not that complicated." Then comes the line that explains the whole system: "If we can't make a decision in five minutes, we can't make it in five months. You know, there's—we're not going to learn enough in the following five months to make up for the fact that we went in deficient in the first place." Deliberation doesn't fix a knowledge deficit. If you weren't already competent to judge the thing, five months of study won't close the gap — it will only manufacture the confidence to act badly. So when the input arrives — a phone call about a business for sale, or a price in a newspaper, a magazine, an annual report, a 10-K — the only thing Buffett is looking for is a "significant differential between price and value." If it's there, "we move right then." "And Charlie and I don't need to talk to each other about it; I mean, we both think the same way and we have generally similar spheres of knowledge." Charlie Munger then names the mechanism directly: "The answer to your question is we can make a lot of decisions about a lot of things very fast and very easily, and we're unusual in that respect. And the reason we're able to do that is there's such an enormous other lot of things that we won't allow ourselves to think about at all. It's just that simple." He gives his own example: "I have a little phrase when people make pitches to me, and about halfway through the first sentence I say, 'We don't do startups; they don't exist.' Well, if you blot out startups, there's a whole layer of complexity that goes out of your life." And he confirms this is a system, not a one-off: "And we've got other little 'blotter out' systems, and using those we finally find out that what remains is still a pretty large territory that we can handle." Buffett closes with the part most people get backwards: "We waste—I would say we waste a lot of time, but we waste it on things we want to waste our time on. And then we're very selective about that, and then we're good at it." The five-minute decision is not actually made in five minutes. Source: 2008 Berkshire Hathaway Annual Meeting
4:52

Warren Buffett: "If we can't make a decision in five minutes, we can't make it in five months." In a Q&A, a shareholder from Munich, Germany asks Buffett a specific question: how large is the universe of companies whose intrinsic value he carries in his head — the ones he could act on within a day or two if the market offered an attractive price? Buffett doesn't give a number. He reframes the question entirely. Speed, he explains, doesn't come from knowing more. It comes from refusing to think about most things at all. "Our immediate decision is whether we can figure out what's being offered to us or not. I mean, there's a go no-go signal." That signal fires almost immediately: "Charlie and I are often thought to be rude when we think we're just being polite and not wasting the other person's time. So, as they start mid-sentence in their first conversation with us, we just say, 'Forget it.'" He continues: "We know very, very, very early in the conversation whether somebody's talking about something that there's any chance is actionable by us, and we don't worry about the ones we miss." The filter isn't about the quality of the opportunity. It's about whether Buffett is equipped to judge it: "We want to make sure that we don't waste any time thinking about things that, when we got all through thinking about them, we're not going to know enough to make the decision on. So we just rule those out, and that rules a lot of things out." What survives that filter gets decided on immediately: "So we make decisions—we can make a decision in five minutes very easily. I mean, it just is not that complicated." Then comes the line that explains the whole system: "If we can't make a decision in five minutes, we can't make it in five months. You know, there's—we're not going to learn enough in the following five months to make up for the fact that we went in deficient in the first place." Deliberation doesn't fix a knowledge deficit. If you weren't already competent to judge the thing, five months of study won't close the gap — it will only manufacture the confidence to act badly. So when the input arrives — a phone call about a business for sale, or a price in a newspaper, a magazine, an annual report, a 10-K — the only thing Buffett is looking for is a "significant differential between price and value." If it's there, "we move right then." "And Charlie and I don't need to talk to each other about it; I mean, we both think the same way and we have generally similar spheres of knowledge." Charlie Munger then names the mechanism directly: "The answer to your question is we can make a lot of decisions about a lot of things very fast and very easily, and we're unusual in that respect. And the reason we're able to do that is there's such an enormous other lot of things that we won't allow ourselves to think about at all. It's just that simple." He gives his own example: "I have a little phrase when people make pitches to me, and about halfway through the first sentence I say, 'We don't do startups; they don't exist.' Well, if you blot out startups, there's a whole layer of complexity that goes out of your life." And he confirms this is a system, not a one-off: "And we've got other little 'blotter out' systems, and using those we finally find out that what remains is still a pretty large territory that we can handle." Buffett closes with the part most people get backwards: "We waste—I would say we waste a lot of time, but we waste it on things we want to waste our time on. And then we're very selective about that, and then we're good at it." The five-minute decision is not actually made in five minutes. Source: 2008 Berkshire Hathaway Annual Meeting

Finance Nerd

54,443 views • 1 month ago

"To be honest, there's a lot of things that go on that you guys don't see. A lot of things we talk about, a lot of things we do that we don't even say publicly. For a reason - because we want to keep everything in-house. We're all we got, no matter what" - Jalen Brunson
0:43

"To be honest, there's a lot of things that go on that you guys don't see. A lot of things we talk about, a lot of things we do that we don't even say publicly. For a reason - because we want to keep everything in-house. We're all we got, no matter what" - Jalen Brunson

SNY Knicks

236,785 views • 5 months ago

Alexandr Wang, Meta's Chief AI Officer, on why Meta can no longer simply open-source its frontier model: As part of standing up Meta Superintelligence Labs, the team rewrote its internal risk doctrine. "One of the things that we did as part of Meta Superintelligence Labs is we updated our what we call our advanced AI scaling framework which is really our view of what are the risks that we see in developing these very powerful models and how do we want to handle those risks as we see them in early testing." They then ran their frontier model through it, and published what came back. "We published a lot of what we saw in the process of training Muark in our preparedness report and some of the things that we saw is that it actually triggered some high risk areas in the course of early training particularly around biorisk but also a number of the risks were elevated." The trigger came during early training, well before launch or red-teaming. Biorisk was the standout, with several other categories rising alongside it. Alexandr Wang is clear this isn't specific to Meta: "This is something I think the entire industry has seen as the models have improved pretty dramatically over the past year so we certainly aren't the only ones to see a host of these risks show up as we scaled up the models and as we sort of kept pushing the frontier of research." Which brings him to the real fork in the road: the difference between shipping a model inside a product and handing out the weights. "When we launched a model like New Spark in a product, we have a lot of ways to mitigate some of these risks and ensure that we're able to launch it in a safe and responsible way. It's much harder to do that when you open source a model and people can use that model in all sorts of contexts that we may not have full understanding of." A product is a controlled surface. You can filter, monitor, rate-limit, patch and revoke. An open-weights release is a one-way door: once the file is out, the deployment context and the mitigations both stop being yours. So Meta is building something different for release: "So we're in the process right now of developing models that we believe are fit and safe to be open source while still maintaining as much of the performance capabilities as possible."
1:27

Alexandr Wang, Meta's Chief AI Officer, on why Meta can no longer simply open-source its frontier model: As part of standing up Meta Superintelligence Labs, the team rewrote its internal risk doctrine. "One of the things that we did as part of Meta Superintelligence Labs is we updated our what we call our advanced AI scaling framework which is really our view of what are the risks that we see in developing these very powerful models and how do we want to handle those risks as we see them in early testing." They then ran their frontier model through it, and published what came back. "We published a lot of what we saw in the process of training Muark in our preparedness report and some of the things that we saw is that it actually triggered some high risk areas in the course of early training particularly around biorisk but also a number of the risks were elevated." The trigger came during early training, well before launch or red-teaming. Biorisk was the standout, with several other categories rising alongside it. Alexandr Wang is clear this isn't specific to Meta: "This is something I think the entire industry has seen as the models have improved pretty dramatically over the past year so we certainly aren't the only ones to see a host of these risks show up as we scaled up the models and as we sort of kept pushing the frontier of research." Which brings him to the real fork in the road: the difference between shipping a model inside a product and handing out the weights. "When we launched a model like New Spark in a product, we have a lot of ways to mitigate some of these risks and ensure that we're able to launch it in a safe and responsible way. It's much harder to do that when you open source a model and people can use that model in all sorts of contexts that we may not have full understanding of." A product is a controlled surface. You can filter, monitor, rate-limit, patch and revoke. An open-weights release is a one-way door: once the file is out, the deployment context and the mitigations both stop being yours. So Meta is building something different for release: "So we're in the process right now of developing models that we believe are fit and safe to be open source while still maintaining as much of the performance capabilities as possible."

Big Brain AI

16,374 views • 1 month ago

Frost: After walking out of this we all have a lot of new questions that we have to raise as it relates to the reasoning on why Donald Trump had a falling out with Jeffrey Epstein in the first place.
0:43

Frost: After walking out of this we all have a lot of new questions that we have to raise as it relates to the reasoning on why Donald Trump had a falling out with Jeffrey Epstein in the first place.

Acyn

32,448 views • 6 months ago

Listen to JD Vance finally acknowledge that the inflation crisis in America is real and that he and Trump have a lot of work to do. "Well, I guess my message to Americans is we know that we have a lot of work to do in order to deliver on the prosperity that the American people deserve. The President is hyper-aware of this. I'm hyper-aware of this. We talk about it all the time. We know that the American people inherited a very bad inflation crisis. Yes, the inflation number last month was not great."
0:32

Listen to JD Vance finally acknowledge that the inflation crisis in America is real and that he and Trump have a lot of work to do. "Well, I guess my message to Americans is we know that we have a lot of work to do in order to deliver on the prosperity that the American people deserve. The President is hyper-aware of this. I'm hyper-aware of this. We talk about it all the time. We know that the American people inherited a very bad inflation crisis. Yes, the inflation number last month was not great."

Ed Krassenstein

32,101 views • 4 months ago

Demis on why world models are his longest standing passion and explains benefits vs. language models: ▫️ “I think language models are able to understand a lot about the world. More than we expected because language is actually probably richer than we thought. But there's still a lot about the spatial dynamics of the world, spatial awareness and the physical context we're in — and how that works mechanically — that is hard to describe in words and isn't generally described in corpuses of words. Alot of this is allied to learning from experience. There's a lot of things which you can't really describe something. You have to just experience it. Maybe the senses and so on are very hard to put into words. Whether that's motor angles and smell and these kinds of senses, it's very difficult to describe that in any kind of language.”▫️
2:32

Demis on why world models are his longest standing passion and explains benefits vs. language models: ▫️ “I think language models are able to understand a lot about the world. More than we expected because language is actually probably richer than we thought. But there's still a lot about the spatial dynamics of the world, spatial awareness and the physical context we're in — and how that works mechanically — that is hard to describe in words and isn't generally described in corpuses of words. Alot of this is allied to learning from experience. There's a lot of things which you can't really describe something. You have to just experience it. Maybe the senses and so on are very hard to put into words. Whether that's motor angles and smell and these kinds of senses, it's very difficult to describe that in any kind of language.”▫️

Bearly AI

78,545 views • 3 months ago

Elon Musk: We could have saved a lot of tragedy in LA. “We really need to have fire breaks, and we need to clear the brush back away from houses. And we need to make sure the reservoirs are full. These are all obvious things. But due to a bunch of environmental rulings, you can't actually do that in California. They're not allowed to do the fire breaks, and they're not allowed to push the brush back away from houses because it might hurt some red-legged frog or something like that. It's like some sort of creature, usually a creature you've never heard of, that is preventing this from occurring. There's this fish called the smelt, for example. We have far more freshwater runoff into the ocean than we should on the theory that it helps this one little fish that likes a slightly briny freshwater salt water mix. And if we keep more fresh water, then the smelt fish will not be happy. But there's no actual evidence that the smelt fish is going to be unhappy if we keep a bit more fresh water. So, we should keep more fresh water, keep the reservoirs full, and just have some sensible fire breaks and move the brush away from houses. We'd have saved a lot of trouble, a lot of tragedy in LA if we'd done that.”
1:23

Elon Musk: We could have saved a lot of tragedy in LA. “We really need to have fire breaks, and we need to clear the brush back away from houses. And we need to make sure the reservoirs are full. These are all obvious things. But due to a bunch of environmental rulings, you can't actually do that in California. They're not allowed to do the fire breaks, and they're not allowed to push the brush back away from houses because it might hurt some red-legged frog or something like that. It's like some sort of creature, usually a creature you've never heard of, that is preventing this from occurring. There's this fish called the smelt, for example. We have far more freshwater runoff into the ocean than we should on the theory that it helps this one little fish that likes a slightly briny freshwater salt water mix. And if we keep more fresh water, then the smelt fish will not be happy. But there's no actual evidence that the smelt fish is going to be unhappy if we keep a bit more fresh water. So, we should keep more fresh water, keep the reservoirs full, and just have some sensible fire breaks and move the brush away from houses. We'd have saved a lot of trouble, a lot of tragedy in LA if we'd done that.”

ELON CLIPS

2,798,498 views • 1 year ago

David Stearns was asked how he approaches the Mets losing long-tenured, popular players: "That is something, broadly, that we talk about a lot. We have a lot of those conversations. We've had them throughout this offseason. We had them throughout last offseason, as well, when we were faced with some decisions. It is part of it. I've said this before, I can't tell you I know exactly how to weigh that. I think we do our best to weigh the full impact of any player on our team, on our organization, and we make the best decision we can."
0:45

David Stearns was asked how he approaches the Mets losing long-tenured, popular players: "That is something, broadly, that we talk about a lot. We have a lot of those conversations. We've had them throughout this offseason. We had them throughout last offseason, as well, when we were faced with some decisions. It is part of it. I've said this before, I can't tell you I know exactly how to weigh that. I think we do our best to weigh the full impact of any player on our team, on our organization, and we make the best decision we can."

SNY Mets

61,721 views • 9 months ago

Jensen to AI Leaders: “We have to be far more thoughtful” when communicating to the public Jensen Huang: “(AI) is not a biological being. It is not alien. It is not conscious. It is computer software.” “We say things like, ‘We don't understand it at all.’ It is not true. We understand a lot of things about this technology.” Chamath: “If you were in the seat in the boardroom of Anthropic over that whole scuttlebutt with the Department of War, what do you think you would've told Dario and that team to do, maybe, differently to try to change some of this outcome and some of this perception?” Jensen: “The first thing that I would say about Anthropic is, first of all, the technology is incredible. We are a large consumer of Anthropic technology.” “The desire to warn people about the capability of the technology is also really terrific.” “We just have to make sure that we understand that the world has a spectrum, and that warning is good, scaring is less good because this technology is too important to us.” “I think that it is fine to predict the future, but we need to be a little bit more circumspect. We need to have a little bit more humility, that, in fact, we can't completely predict the future.” “And to say things that are quite extreme, quite catastrophic, that there's no evidence of it happening, could be more damaging than people think.” “And of course we are technology leaders.” “There was a time when nobody listened to us, but now because technology is so important in the social fabric, such an important industry, so important to national security, our words do matter.” “And I think we have to be much more circumspect, we have to be more moderate, we have to be more balanced, we have to be far more thoughtful.”
1:27

Jensen to AI Leaders: “We have to be far more thoughtful” when communicating to the public Jensen Huang: “(AI) is not a biological being. It is not alien. It is not conscious. It is computer software.” “We say things like, ‘We don't understand it at all.’ It is not true. We understand a lot of things about this technology.” Chamath: “If you were in the seat in the boardroom of Anthropic over that whole scuttlebutt with the Department of War, what do you think you would've told Dario and that team to do, maybe, differently to try to change some of this outcome and some of this perception?” Jensen: “The first thing that I would say about Anthropic is, first of all, the technology is incredible. We are a large consumer of Anthropic technology.” “The desire to warn people about the capability of the technology is also really terrific.” “We just have to make sure that we understand that the world has a spectrum, and that warning is good, scaring is less good because this technology is too important to us.” “I think that it is fine to predict the future, but we need to be a little bit more circumspect. We need to have a little bit more humility, that, in fact, we can't completely predict the future.” “And to say things that are quite extreme, quite catastrophic, that there's no evidence of it happening, could be more damaging than people think.” “And of course we are technology leaders.” “There was a time when nobody listened to us, but now because technology is so important in the social fabric, such an important industry, so important to national security, our words do matter.” “And I think we have to be much more circumspect, we have to be more moderate, we have to be more balanced, we have to be far more thoughtful.”

The All-In Podcast

57,581 views • 5 months ago

"This is the first time in over a decade that we see even the glimmer of an opportunity to change conditions. A lot of that will depend on us but a lot of that will depend on them, and it also will depend on the rest of Venezuelan society ... We have to be much further along in six months. We expect to be further along." — Secretary Marco Rubio
0:46

"This is the first time in over a decade that we see even the glimmer of an opportunity to change conditions. A lot of that will depend on us but a lot of that will depend on them, and it also will depend on the rest of Venezuelan society ... We have to be much further along in six months. We expect to be further along." — Secretary Marco Rubio

Department of State

127,257 views • 7 months ago

This is my first long conversation in a very, very long time. The idea was to talk about all the things that make Zerodha what it is today, because a lot of people don’t really know how differently Zerodha has been built compared to most other companies. We’ve never raised external capital. We don’t advertise. We don’t have a sales team with sales targets. We don’t do much marketing. We don’t push people to trade or invest. And we have a relatively small team that has remained at roughly 1,000 people since around 2018–19, even though the business has grown five or six times since then. In this conversation, I spoke about all these things and more: how we think about growth, risk, competition, and the future of Zerodha. This is also the beginning of a new series where we want to shine a spotlight on the people who have made Zerodha what it is today. A lot of them aren’t very well known, so the idea is to put them on the spot and tell their stories. Check out the full conversation on the Zerodha YouTube channel. The link is in the comments.
1:07

This is my first long conversation in a very, very long time. The idea was to talk about all the things that make Zerodha what it is today, because a lot of people don’t really know how differently Zerodha has been built compared to most other companies. We’ve never raised external capital. We don’t advertise. We don’t have a sales team with sales targets. We don’t do much marketing. We don’t push people to trade or invest. And we have a relatively small team that has remained at roughly 1,000 people since around 2018–19, even though the business has grown five or six times since then. In this conversation, I spoke about all these things and more: how we think about growth, risk, competition, and the future of Zerodha. This is also the beginning of a new series where we want to shine a spotlight on the people who have made Zerodha what it is today. A lot of them aren’t very well known, so the idea is to put them on the spot and tell their stories. Check out the full conversation on the Zerodha YouTube channel. The link is in the comments.

Nithin Kamath

81,994 views • 1 month ago

Jacob talks about making deliberate choices with his shoots: ‘Another trick that I learned years ago is that I don't shoot very much, so nobody can cut my show the way I don't want them to cut it. There's no excess. "Is there a close up?" and I'm like, "No." And so, I do a lot of that, and it's very deliberate. I mean, the thing about the show when you talk about the money and the resources, the reason that it is stretched to a point, is that we shot, I think, 57 locations in 34 days? So to keep the scope of the show the way we wanted, we had to move fast, and we had to get things done. And so that means there's going to be some one-ers, there's going to be some things, and instead of you seeing that as a limitation, you have to view that as like, “This is a fun opportunity to do something interesting.” We shot that cottage in two days. We had two days to do it, and so it was like, “Okay, here we are doing it.” And we finished early. 🗣️ CSA Presents: Focal Point, May 31, 2026. ▶️ ✍🏼 transcription by Heated Rivalry News
0:55

Jacob talks about making deliberate choices with his shoots: ‘Another trick that I learned years ago is that I don't shoot very much, so nobody can cut my show the way I don't want them to cut it. There's no excess. "Is there a close up?" and I'm like, "No." And so, I do a lot of that, and it's very deliberate. I mean, the thing about the show when you talk about the money and the resources, the reason that it is stretched to a point, is that we shot, I think, 57 locations in 34 days? So to keep the scope of the show the way we wanted, we had to move fast, and we had to get things done. And so that means there's going to be some one-ers, there's going to be some things, and instead of you seeing that as a limitation, you have to view that as like, “This is a fun opportunity to do something interesting.” We shot that cottage in two days. We had two days to do it, and so it was like, “Okay, here we are doing it.” And we finished early. 🗣️ CSA Presents: Focal Point, May 31, 2026. ▶️ ✍🏼 transcription by Heated Rivalry News

Heated Rivalry News

33,607 views • 3 months ago

Demis Hassabis on why world models are his longest standing passion and the benefits vs. language models: ▫️ “I think language models are able to understand a lot about the world. More than we expected because language is actually probably richer than we thought. But there's still a lot about the spatial dynamics of the world, spatial awareness and the physical context we're in — and how that works mechanically — that is hard to describe in words and isn't generally described in corpuses of words. Alot of this is allied to learning from experience. There's a lot of things which you can't really describe something. You have to just experience it. Maybe the senses and so on are very hard to put into words. Whether that's motor angles and smell and these kinds of senses, it's very difficult to describe that in any kind of language.”▫️ *** From the Google DeepMind podcast in December 2025.
2:32

Demis Hassabis on why world models are his longest standing passion and the benefits vs. language models: ▫️ “I think language models are able to understand a lot about the world. More than we expected because language is actually probably richer than we thought. But there's still a lot about the spatial dynamics of the world, spatial awareness and the physical context we're in — and how that works mechanically — that is hard to describe in words and isn't generally described in corpuses of words. Alot of this is allied to learning from experience. There's a lot of things which you can't really describe something. You have to just experience it. Maybe the senses and so on are very hard to put into words. Whether that's motor angles and smell and these kinds of senses, it's very difficult to describe that in any kind of language.”▫️ *** From the Google DeepMind podcast in December 2025.

Bearly AI

65,630 views • 1 month ago

Sam on Kimi, distillation, and open source: “I have always assumed that there are going to be great cheap models in the world, and we better be the greatest and the cheapest. You get a better deal today, at least at a particular latency, using OpenAI's models than Kimi. We distill our own models. That’s how we make smaller, cheaper models. That’s a very good thing to do. There will clearly be an important place for open source models in the world and people that will want their own weights. We have so much usage of our models that we do not need to be a gigantically high-margin business to be able to afford model training. I would rather people not distill from us, for sure. But this is not in my top ten list of worries.”
2:12

Sam on Kimi, distillation, and open source: “I have always assumed that there are going to be great cheap models in the world, and we better be the greatest and the cheapest. You get a better deal today, at least at a particular latency, using OpenAI's models than Kimi. We distill our own models. That’s how we make smaller, cheaper models. That’s a very good thing to do. There will clearly be an important place for open source models in the world and people that will want their own weights. We have so much usage of our models that we do not need to be a gigantically high-margin business to be able to afford model training. I would rather people not distill from us, for sure. But this is not in my top ten list of worries.”

Patrick OShaughnessy

269,663 views • 1 month ago

LatchBio researcher arjun reveals Grok's refusals often come from the model itself while Fable and GPT-5.6 Sol rely more on external safety layers: "There might be input classifiers where when a user puts in a query, the model initially might say, this is a bad query, we can't respond to this. There might also be output classifiers, so after the model outputs a query, you might have an external classifier that says, no, don't show this query, put it in an API block. And there are also distinct types of model reasoning where the model might just say, hey, I cannot answer this question." "We found across the board a lot of Grok's refusals were Grok itself in text saying, I cannot answer this question." "Whereas in models like Fable or even GPT 5.6 Sol, a lot of that's an API block, where there seems to be some sort of additional other part of the system, whether that be a linear probe or some sort of additional external classifier that says, this output cannot be said." "We can discern this by actually just reading the logs that's being outputted." LatchBio Kenny Workman
1:02

LatchBio researcher arjun reveals Grok's refusals often come from the model itself while Fable and GPT-5.6 Sol rely more on external safety layers: "There might be input classifiers where when a user puts in a query, the model initially might say, this is a bad query, we can't respond to this. There might also be output classifiers, so after the model outputs a query, you might have an external classifier that says, no, don't show this query, put it in an API block. And there are also distinct types of model reasoning where the model might just say, hey, I cannot answer this question." "We found across the board a lot of Grok's refusals were Grok itself in text saying, I cannot answer this question." "Whereas in models like Fable or even GPT 5.6 Sol, a lot of that's an API block, where there seems to be some sort of additional other part of the system, whether that be a linear probe or some sort of additional external classifier that says, this output cannot be said." "We can discern this by actually just reading the logs that's being outputted." LatchBio Kenny Workman

MTS

13,719 views • 10 days ago