Loading video...

Video Failed to Load

Go Home

OpenAI made computer use 10x faster in a year. Tibo breaks down what changed under the hood: “It’s not that we made the model sample faster. It’s both the harness and the model.” “How much does it need to think before it can take the next action reliably? That’s...

tbpn's profile picture

TBPN

1,619,596 subscribers

46,224 views • 9 days ago •via X (Twitter)

Science & Technology
LIVE

Anya Rossi• Live Now

Private livecam show

3 Comments

Kyle Jeong's profile picture
Kyle Jeong9 days ago

@thsottiaux

小森's profile picture
小森9 days ago

@thsottiaux Faster is nice. The part I care about is the think-before-click budget—hard approval queue for money/auth, or just shave tokens until it guesses wrong once?

LucasH Sketch's profile picture
LucasH Sketch9 days ago

@thsottiaux a second agent babysitting the first agent is such a funny solution but hey, if it works it works

Related Videos

Jev made our Slack agent 2x faster ⚡️ Our agent can be quite slow because it needs to read skills and figure out which tools to call. We used TypeSafe AI's new model to speed this up by first passing it the prompt and classifying the best skill, tool and params to use before handing it to the agent
1:02

Jev made our Slack agent 2x faster ⚡️ Our agent can be quite slow because it needs to read skills and figure out which tools to call. We used TypeSafe AI's new model to speed this up by first passing it the prompt and classifying the best skill, tool and params to use before handing it to the agent

John Yeo

35,432 views • 20 days ago

#ROSÉ about BLACKPINKOFFICIAL as a group. “We had every decision that we made, we made very carefully. We discussed it a lot […] It’s much more healthier now […] To understand that we are like four individuals who have four different personalities, but all together the things we can do and what we can’t do just to learn that over the course of 9-10ish, almost 10 years. So it’s about time we understand how this works.”
0:48

#ROSÉ about BLACKPINKOFFICIAL as a group. “We had every decision that we made, we made very carefully. We discussed it a lot […] It’s much more healthier now […] To understand that we are like four individuals who have four different personalities, but all together the things we can do and what we can’t do just to learn that over the course of 9-10ish, almost 10 years. So it’s about time we understand how this works.”

Blinks United

294,884 views • 10 months ago

Anthropic Technical Staff Jess Yan says the harness and the model cannot be pulled apart without giving up performance, though she flags her own bias. "I think I am quite biased but I also think that it is impossible to get the maximum possible performance without tying together the harness and the model." "Now the components of the harness and maybe the thickness of the harness will change over time as models get more and more capable." "However, when we test our models and when we are assessing their performance we always have to test it in conjunction with a harness." "And are we going to test it with all the different harnesses of the world? We're going to select the harnesses that we have built." "And so there is an aspect of the necessity of building models is that you have to be testing them with harnesses, and that sort of keeps them paired together."
0:46

Anthropic Technical Staff Jess Yan says the harness and the model cannot be pulled apart without giving up performance, though she flags her own bias. "I think I am quite biased but I also think that it is impossible to get the maximum possible performance without tying together the harness and the model." "Now the components of the harness and maybe the thickness of the harness will change over time as models get more and more capable." "However, when we test our models and when we are assessing their performance we always have to test it in conjunction with a harness." "And are we going to test it with all the different harnesses of the world? We're going to select the harnesses that we have built." "And so there is an aspect of the necessity of building models is that you have to be testing them with harnesses, and that sort of keeps them paired together."

Fireside Alpha

584,074 views • 2 months ago

OpenAI’s Tibo explains how dots work and how OpenAI has been deploying specialist dots internally: “The agent never stops. Occasionally it sleeps. But it’s always online, always doing proactive research, thinking about your goals, thinking about your preferences.” “You just give your dot a little bit of feedback. It learns.” “A couple days in, you’re like, ‘Wow, I can’t really quite imagine doing work without this thing.’” “You don’t have to be super technical but just kind of still feel the power of it.” Inside OpenAI, specialist dots are already being provisioned with their own identities, permissions, machines, and roles: “We have them running on Mac minis.” “We provision them exactly like employees. We mint a new little identity, and then it’s got this IAM and the roles.” “We give them a very premium machine as well because the model is incredibly capable under the hood.” “So your dot can just use a computer just like you would.” “We have these set up inside of OpenAI. They do all sorts of things, including procurement, for example.”
5:02

OpenAI’s Tibo explains how dots work and how OpenAI has been deploying specialist dots internally: “The agent never stops. Occasionally it sleeps. But it’s always online, always doing proactive research, thinking about your goals, thinking about your preferences.” “You just give your dot a little bit of feedback. It learns.” “A couple days in, you’re like, ‘Wow, I can’t really quite imagine doing work without this thing.’” “You don’t have to be super technical but just kind of still feel the power of it.” Inside OpenAI, specialist dots are already being provisioned with their own identities, permissions, machines, and roles: “We have them running on Mac minis.” “We provision them exactly like employees. We mint a new little identity, and then it’s got this IAM and the roles.” “We give them a very premium machine as well because the model is incredibly capable under the hood.” “So your dot can just use a computer just like you would.” “We have these set up inside of OpenAI. They do all sorts of things, including procurement, for example.”

TBPN

41,491 views • 9 days ago

Miller expands that ‘we need to take on, uproot and disembed the Zionist movement in all of our western countries’. And, ‘It’s not just…that we think about how we support the resistance in Palestine, it’s that we think about how we practically do it materially in the UK’ 🎥
1:33

Miller expands that ‘we need to take on, uproot and disembed the Zionist movement in all of our western countries’. And, ‘It’s not just…that we think about how we support the resistance in Palestine, it’s that we think about how we practically do it materially in the UK’ 🎥

Danny Morris

46,488 views • 1 year ago

🗣️ Mikel Arteta on Arsenal having “the best midfield in the world”: "It’s the best midfield for us, and that’s for sure. “That’s what we have to make it, that everybody complements the player next to them with their qualities. “We have this variability, this diversity with the qualities that we need in relation to the opponent. “Today again, you’ve seen it. We changed it and we brought other qualities in different moments of the game. “That’s a really, really strong thing to have.”
0:27

🗣️ Mikel Arteta on Arsenal having “the best midfield in the world”: "It’s the best midfield for us, and that’s for sure. “That’s what we have to make it, that everybody complements the player next to them with their qualities. “We have this variability, this diversity with the qualities that we need in relation to the opponent. “Today again, you’ve seen it. We changed it and we brought other qualities in different moments of the game. “That’s a really, really strong thing to have.”

Connor Humm

27,471 views • 1 month ago

Arteta: “Throughout this journey, we have made some massive steps, and we have accomplished a lot of things that, in my opinion, have a lot of value, but at the end of the day, we are here to win major trophies. “That was the ultimate goal, and we were very close. And in three occasions, we fell short at the end, and that was very painful. But I think that's what has driven all of us to find new ways to show what we are made of. That's what I said, the manner that we've done it, it makes it even better.”
0:35

Arteta: “Throughout this journey, we have made some massive steps, and we have accomplished a lot of things that, in my opinion, have a lot of value, but at the end of the day, we are here to win major trophies. “That was the ultimate goal, and we were very close. And in three occasions, we fell short at the end, and that was very painful. But I think that's what has driven all of us to find new ways to show what we are made of. That's what I said, the manner that we've done it, it makes it even better.”

Connor Humm

14,155 views • 4 months ago

Astra began to saturate a computer-use eval we haven’t even released yet. Astra was not only able to kill several hostile mobs in Minecraft, but it made it all the way to building a Nether portal in less than three hours. We believe that if we ran it longer, it would beat the game. Astra was able to do things like tower up to kill a horde of zombies, trap a skeleton inside a hole to dodge its arrows, craft a boat, and navigate back to its base. Here is the full 3 hour play through. All of this was without any specialized harness: pure computer use.
3:01:05

Astra began to saturate a computer-use eval we haven’t even released yet. Astra was not only able to kill several hostile mobs in Minecraft, but it made it all the way to building a Nether portal in less than three hours. We believe that if we ran it longer, it would beat the game. Astra was able to do things like tower up to kill a horde of zombies, trap a skeleton inside a hole to dodge its arrows, craft a boat, and navigate back to its base. Here is the full 3 hour play through. All of this was without any specialized harness: pure computer use.

Vals AI

37,552 views • 1 month ago

This week we’re getting GPT 5.5 and I couldn’t be more excited to see the first new pre train base in a while Before Spud releases, here’s the last video you need to watch - Greg Brockman perfectly indicated to us 3 important things that perfectly encapsulate the lineage of OpenAI and the mission. 1.This is the culmination of 2 years of research and is a fresh base pre-train that we haven’t gotten for a long time. 2.This is an “exciting release” and it will bring new capabilities (sort of a given but still very grateful nonetheless) & that it is AN EARLY VERSION of what we have coming - meaning they have a lot of RL to do to make this model even better since it’s a fresh pre-train. 3.THIS IS THE BEGINNING OF THE FLYWHEEL. As he said toward the end, “we have this engine of progress that just moves faster and faster and Spud is just one step along the way.” Depending on the capabilities we see this week from Spud, we may be able to map out if AGI will be sooner rather than later, judging by how much recursive work can exist within this model and the harness. In other words, the flywheel.
0:42

This week we’re getting GPT 5.5 and I couldn’t be more excited to see the first new pre train base in a while Before Spud releases, here’s the last video you need to watch - Greg Brockman perfectly indicated to us 3 important things that perfectly encapsulate the lineage of OpenAI and the mission. 1.This is the culmination of 2 years of research and is a fresh base pre-train that we haven’t gotten for a long time. 2.This is an “exciting release” and it will bring new capabilities (sort of a given but still very grateful nonetheless) & that it is AN EARLY VERSION of what we have coming - meaning they have a lot of RL to do to make this model even better since it’s a fresh pre-train. 3.THIS IS THE BEGINNING OF THE FLYWHEEL. As he said toward the end, “we have this engine of progress that just moves faster and faster and Spud is just one step along the way.” Depending on the capabilities we see this week from Spud, we may be able to map out if AGI will be sooner rather than later, judging by how much recursive work can exist within this model and the harness. In other words, the flywheel.

Chris

103,198 views • 5 months ago

Today is Day 11 of 12 Days of Agent Two. Agent Two isn't just Claude under the hood - we get asked this a lot. The truth is - while we use all the state of the art models out there, the Agent's real intelligence comes from its harness: a domain-specific layer built from continuous creative research that shapes how every model is actually used - more about this in the video. But this harness is what makes Agent Two frontier intelligence for creative work - at 4x the speed, and half the price.
5:00

Today is Day 11 of 12 Days of Agent Two. Agent Two isn't just Claude under the hood - we get asked this a lot. The truth is - while we use all the state of the art models out there, the Agent's real intelligence comes from its harness: a domain-specific layer built from continuous creative research that shapes how every model is actually used - more about this in the video. But this harness is what makes Agent Two frontier intelligence for creative work - at 4x the speed, and half the price.

Invideo

2,561,777 views • 1 month ago

We just made Astra's computer use 2.5x faster with Stagehand. Astra works by executing code against the a11y tree, we built a translator that turns Astra's playwright commands into Stagehand. Astra chose to batch the entire logo into one command and one-shotted it.
0:39

We just made Astra's computer use 2.5x faster with Stagehand. Astra works by executing code against the a11y tree, we built a translator that turns Astra's playwright commands into Stagehand. Astra chose to batch the entire logo into one command and one-shotted it.

Kyle Jeong

24,181 views • 1 month ago

Same Doge. New beat. Five years ago, Elon Musk challenged the Dogecoin community to make it 10x faster, increase block size 10x, and cut fees 100x. It didn’t look possible on the existing proof-of-work stack. At the same time, at Metallicus and Metal Blockchain, we had begun the work of building blockchain technology to integrate into banks, credit unions, and financial services. I could see where the future was headed, and the vision was clear. It took a few years. We didn’t just make it 10x faster. We made it 300x faster. Throughput no longer depends on block size, and fees are fully customizable. This isn’t Dogecoin Core. It’s a separate payment layer that keeps the same UTXO model as Dogecoin and Bitcoin, under a proof-of-stake consensus called Snowman. Built for X, and for any fintech that wants to scale Dogecoin and Bitcoin. And we’re not finished. We opened the door. On the other side is the blockchainification of finance as we know it. GitHub: Live alpha on mainnet:
2:50

Same Doge. New beat. Five years ago, Elon Musk challenged the Dogecoin community to make it 10x faster, increase block size 10x, and cut fees 100x. It didn’t look possible on the existing proof-of-work stack. At the same time, at Metallicus and Metal Blockchain, we had begun the work of building blockchain technology to integrate into banks, credit unions, and financial services. I could see where the future was headed, and the vision was clear. It took a few years. We didn’t just make it 10x faster. We made it 300x faster. Throughput no longer depends on block size, and fees are fully customizable. This isn’t Dogecoin Core. It’s a separate payment layer that keeps the same UTXO model as Dogecoin and Bitcoin, under a proof-of-stake consensus called Snowman. Built for X, and for any fintech that wants to scale Dogecoin and Bitcoin. And we’re not finished. We opened the door. On the other side is the blockchainification of finance as we know it. GitHub: Live alpha on mainnet:

Marshall Hayner

39,855 views • 2 days ago

Karl Popper on How We Gain Knowledge: ––Karl Popper: "People think, usually, that we acquire knowledge by opening our eyes and our ears and let the sensations stream into us, and they believe then that we record this like a camera. In my opinion, if we wish to get knowledge, we have to have a problem. It has to be knowledge of something. We have to find out something. We don't have to wait for information to stream into us, but we have to be inquisitive if we want to get knowledge. If we were passive, we would gain a confused mass of sensations or something like that, which we would hardly be able to understand and to convert into what one may call knowledge. Quite apart from that, perception is not really, in my opinion, the main source of our knowledge. The role of perception is to inform us about a momentary situation in our environment. But we couldn't really interpret our perceptions without knowing much more about our environment, namely, we know whether we are in a house or whether we are in a glacier. So we have two kinds of knowledge: this wider knowledge of a frame in which we orientate ourselves, and the momentary perception which gives us information about the situation at that particular moment. And it is only this situation in which we can use our perception. So we have theoretical knowledge and, if you like, the momentary practical challenge to our theoretical knowledge. And here comes perception in."
3:03

Karl Popper on How We Gain Knowledge: ––Karl Popper: "People think, usually, that we acquire knowledge by opening our eyes and our ears and let the sensations stream into us, and they believe then that we record this like a camera. In my opinion, if we wish to get knowledge, we have to have a problem. It has to be knowledge of something. We have to find out something. We don't have to wait for information to stream into us, but we have to be inquisitive if we want to get knowledge. If we were passive, we would gain a confused mass of sensations or something like that, which we would hardly be able to understand and to convert into what one may call knowledge. Quite apart from that, perception is not really, in my opinion, the main source of our knowledge. The role of perception is to inform us about a momentary situation in our environment. But we couldn't really interpret our perceptions without knowing much more about our environment, namely, we know whether we are in a house or whether we are in a glacier. So we have two kinds of knowledge: this wider knowledge of a frame in which we orientate ourselves, and the momentary perception which gives us information about the situation at that particular moment. And it is only this situation in which we can use our perception. So we have theoretical knowledge and, if you like, the momentary practical challenge to our theoretical knowledge. And here comes perception in."

Karl Popper Quotes

115,377 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 • 2 months ago

.Senator Amy Klobuchar on rising political violence in the US: “This has gotten totally out of hand... We actually put out a joint statement that said we speak with one voice in condemning political violence. I think we need to see more of this across our country. People need to call people out. Some people need to look in the mirror and say, hey, I have got to stop this or stop my colleagues from doing this, because it makes it much worse. We need to bring the tone down, and we also need to stand up when people do bad things.”
2:24

.Senator Amy Klobuchar on rising political violence in the US: “This has gotten totally out of hand... We actually put out a joint statement that said we speak with one voice in condemning political violence. I think we need to see more of this across our country. People need to call people out. Some people need to look in the mirror and say, hey, I have got to stop this or stop my colleagues from doing this, because it makes it much worse. We need to bring the tone down, and we also need to stand up when people do bad things.”

State of the Union

17,915 views • 1 year ago

Love. It’s a word we use to describe how we feel about many things. And we use it a lot! But what does it really mean to love? Watch this video for the surprising answer.
0:59

Love. It’s a word we use to describe how we feel about many things. And we use it a lot! But what does it really mean to love? Watch this video for the surprising answer.

Right From The Heart

315,663 views • 2 years ago

Watched this speech about 10x today. I beg you, if you’re Panjabi, use 5 mins to watch this. Learn. We sometimes automatically assume as panjabis we have this value set, we see things in the same light. It’s not true. We need to decide which part of Panjab we represent.
4:36

Watched this speech about 10x today. I beg you, if you’re Panjabi, use 5 mins to watch this. Learn. We sometimes automatically assume as panjabis we have this value set, we see things in the same light. It’s not true. We need to decide which part of Panjab we represent.

Jassa

26,257 views • 1 year ago

Anthropic engineer: “At Anthropic, we don’t just write prompts anymore. We build harnesses and loops around agents.” In 30 minutes, she shows how Anthropic builds agent systems with memory, versioning, concurrency, and feedback loops. If this cost $400, people would call it one of the best agent engineering courses of the year. It’s free. Watch it, then use the step-by-step Harness Engineering guide below to build your own.
29:46

Anthropic engineer: “At Anthropic, we don’t just write prompts anymore. We build harnesses and loops around agents.” In 30 minutes, she shows how Anthropic builds agent systems with memory, versioning, concurrency, and feedback loops. If this cost $400, people would call it one of the best agent engineering courses of the year. It’s free. Watch it, then use the step-by-step Harness Engineering guide below to build your own.

Ichigo

17,886 views • 1 month ago