Phil Trubey
@PTrubey • 13,935 subscribers
Looking for AI startups with fundamental technology.
Videos
This morning at NeurIPS, Rich Sutton reminded us that we need continual learning to reach AGI. This afternoon, Ali Behrouz presented a Google poster paper, Nested Learning, which provides new ideas on the path to continual learning. I recorded the 40 minute talk as it might be useful for some researchers in the audience. For the rest of us, I subscribe to Andrej Karpathy's suspicion that it will take a 5-10 papers like this to move us to AGI from where we are now, just like it took about 10 papers to move from 2012's AlexNet to ChatGPT. At the very end, I ask Ali how far along to continual learning this represents. Full paper link below, as well as a YouTube link. ps. sorry about the first 2 minutes of bad audio since there were 2 idiots standing beside me have a conversation right in front of this presenter in a rather packed poster presentation. Honestly, tamp down your egos guys and show come common courtesy!
Phil Trubey235,435 Aufrufe • vor 6 Monaten
NVIDIA CEO Jensen was on fire in his Dwarkesh interview. He makes two incredibly important points in this clip. First, Anthropic and their ilk had better stop scaring everyone that AI is akin to a nuclear bomb. This meme will only hold the US back while the rest of the world (and most importantly China) will pass us by. Second he makes a basic economic argument about worldwide technology dominance. If the US does not allow NVIDIA to foster their ecosystem in China, China will create their own AI tech stack and will eventually crowd out the US’s AI lead everywhere in the world, and eventually in the US itself. This isn’t theoretical. The US conceded manufacturing to China and is desperately trying to claw it back now. 90% of rare earth refining is done in China: raw materials for magnets, meaning motors, meaning drones, meaning military superiority in the end. China is not some backward third world economy that geriatric policy makers might envision. Yes, they are currently behind in advanced chip making, but western lead in that area won’t last. Meanwhile they are ahead in some areas of AI architecture. So the US has a choice. Be confident and compete against China, or eventually cede the AI tech stack to them. Jensen is correct here.
Phil Trubey72,743 Aufrufe • vor 2 Monaten
Brett Adcock, Figure CEO joined our 8 hour(!) live stream where a bunch of us were bird dogging and discussing Figure’s own 8 hour livestream showing a F.03 bot doing a logistics task completely autonomously including shift changes between bots. Here’s my summary of Brett’s remarks. The attached video is just the segment with Brett. We were first introduced to F.03 8 months ago, but Figure has been hard at work on their next version, F.04 which has just completed design lock, so expect to see that new bot sometimes this fall. F.04 was co-designed with the latest Figure AI stack called Helix and was built specifically for data. Brett didn’t explain what that meant, but I suspect it means the bot has many more sensors to enable better training and transfer learning. F.04 will be the biggest leap in performance they’ve had between versions so far which is saying something. Brett is a huge proponent of cross training the bots with many different tasks such that seeming unrelated tasks makes all learned tasks better. He gave an example of the fridge loading training which was topping out at 60% reliability until they trained the same model with kitchen shelving tasks, then they saw the fridge tasks jump to 90% accuracy. As such, they spend almost all their time in pre-training the unified Helix model to ensure they get cross training benefits. Figure will have almost completely localized Figure’s supply chain away from China by next quarter. They build almost everything in-house. Figure does not appear eager to get their bots into the workforce. Brett said they could, today, push thousands of bots into customer hands, and I believe him. But their goal is full general robotics where you can describe a brand new task to a robot, maybe do a one time demonstration, just like you would to a human showing them a new task, and then have the robot do the task. This is the holy grail of AI robotics, and Figure is laser focused on that mission. Brett initially said there was a possibility of achieving it this year, but then guided next couple of years, which I think is much more likely. Personally, I think they’ll need at least a new generation of NVIDIA inference chips to make that leap, and a lot more data gathering, training and hardware development. Brett said their goal with the hardware is “Apple” quality. Ie. Something as well designed and made as any Apple product. While the F.03 hand is clearly performant as shown in the 8 hour livestream, they are building a new hand for the F.04 bot which will be even closer to the full functionality of a human hand. Brett fully believes you need a humanoid hand as close as possible in capability to a human hand, if for no other reason that transfer learning from humans works a lot better when you can exactly mimic what a human does. If the bot can’t do something a human demonstrates, then you’ve just polluted your dataset. By now Figure has built more hands than bot versions (5-6 hands). One of the first hands they tried was a tendon driven hand, and without explaining why, Brett said that was a dead end. Their hands now have all actuators in the hand itself, and are clearly already robust. Brett said he just sat through a 100 page powerpoint design review of the latest hand - that’s how complicated it is. Brett’s other AI company, Hark Labs, has developed a conversational voice model which is installed now in the Figure bots roaming the office. Being able to converse back and forth with a Figure bot is now a thing and will get better over time. All in all, I came away from this segment even more bullish on Figure.
Phil Trubey33,861 Aufrufe • vor 1 Monat
This is why Elon is singular. He has recently spent many weekends working with the AI5 chip team to get the project on track. Designing this chip requires deep knowledge of leading edge process node chip design and leading edge AI algorithms, neither of which Elon has deep experience in. But he learns fast and can ask the right engineering level technical questions to be able to make quick decisions. And along the way, he axed the Dojo program knowing that AI5 was more important. In 99% of other companies, both programs would have chugged along into eventual irrelevance.
Phil Trubey135,536 Aufrufe • vor 7 Monaten
I keep seeing this interview being recommended in my timeline, but I had already seen Dylan being interviewed recently so didn’t think this would add much. I was wrong. This is a banger 40 minute interview chock full of interesting insights. This clip is about robotics, but there is tons of other insights about the entire AI industry.
Phil Trubey22,715 Aufrufe • vor 2 Monaten
Elon about Terafab: “We’re not going to do just conventional compute, I think there’s some very interesting new physics” Very interesting! NVIDIA style chips are just one way of doing AI compute. Is Elon talking about analog AI? Neuromorphic chips? I hope it’s something very ambitious. I think we’re stuck in a local minima in AI hardware and I’m excited to see a big ambitious R&D effort to try and break out. Here’s the meat of the Terafab announcement where he describes what makes it different from all other AI hardware efforts.
Phil Trubey26,175 Aufrufe • vor 3 Monaten
Elon on Joe Rogan confirms the Robovan is coming. I was getting worried because we haven't seen any indication of its production. However, we see hints in Zanegler drone flyovers that it might be manufactured at the Tesla semi cab factory in Nevada, which is due to come on line in a couple of months. This is Tesla's answer to the large SUV market segment they've ignored. A stripped down Robovan will almost certainly be made available for custom shops to turn them into weekend getaway vehicles, campers, ski party busses, mobile offices, etc. Some people will want to own and customize them to their liking, while others will be content to rent them, just like the camper industry today.
Phil Trubey48,241 Aufrufe • vor 7 Monaten
Following my post about Dojo's TTPoE communications protocol that replaces TCP, I want to learn about the Etherloop protocol that Tesla uses in the Cybertruck. Turns out it's the 2nd protocol that Tesla invented, and it also rides on top of Ethernet, but not the same Ethernet that we're all used to. And while researching this I came across a bunch of other related information. It's too complex to put in X posts, so I made and narrated a 24 minute Powerpoint (sorry!) presentation about it all. I hope you enjoy it, or at the very least, it might help cure your insomnia. YouTube link below as well.
Phil Trubey53,924 Aufrufe • vor 1 Jahr
Keine weiteren Inhalte verfügbar