Alphachip
βI've done some work on AlphaChip as one example, helping with placement and routing. Very great Nature paper, by the way.β
What the smart people are recommending. 8874 books, tools, and products endorsed by the thinkers absorb.md tracks. Ranked by how many times each has been recommended across compiled podcasts, papers, posts, and tweets.
βI've done some work on AlphaChip as one example, helping with placement and routing. Very great Nature paper, by the way.β
βI have a simple collab notebook for just extracting emails from the Gmail API... I can share that collab notebook if anybody wants to mess around with that.β
βDid you see the demo of the Gemini? ...Unbelievably fast. Yeah. I'm really excited... I'll get you on the list for access to that.β
βWith the Veo 3, it's like kind of opens up a new dimension... it definitely opens up the space of possibilities for usβ
βThen we developed a program based on reinforcement learning called NVCell. I think we're up to NVCell 2 or 3 now. It's overnight on one GPU. The results are actually better than the human designs...β
βthere is parallel work from DeepMind called AI co-scientist which demonstrates this very clearly that if you propose ideas in natural language then you can get language models to provide meaningful crβ¦β
βThen we have this program called PrefixRL, which applies reinforcement learning to a really age-old problem... It comes up with totally bizarre designs that no human would ever come up with, but they'β¦β
βI'm not sure if you used deep research. Deep research is this new trend in AI where Chad has it, Claude has it, Perplexi, they all have it.β
βvarious things with AI programs like cling and luma and midjourney and um a whole bunch of run re runway and othersβ
βAt a more holistic stage, for a number of years we've had a series of LLMs that we called Chip NeMo and Bug NeMo... Now you have this LLM that's actually very smart about GPU design.β
βAt a more holistic stage, for a number of years we've had a series of LLMs that we called Chip NeMo and Bug NeMo... One of the biggest gains of this is when you have a junior designer.β
βYou have to have cuDNN installed on your system. The cuDNN installation instructions with apt-get will grab the default set of cuDNN packages.β
βMake sure you install MPI and NCCL... For NCCL follow the instructions from the official websiteβ
βan article came out that I loved by the zed team the high performance text editor with AI native uh capabilities or code editor.β
βfor evals, we use a product called brain trust for example and we want to make sure that every product that we don't build but you still need in the AI world is in our marketplace.β
βInstall the Vercel plugin for your favorite coding agentβ
βyou can expect integrations like uh browser base and browser use. They're building browser infrastructure for agents.β
βOpen claw basically put into the popular consciousness what an AI agent can do... open claw is open but it formulates it structures a type of computing model that is basically reinventing computing alβ¦β
βcertainly as a patient you should be using second opinion. But but we don't know if it introduces bias.β
βjust take 03 and ask it a bunch of hard questions and open up its thinking and you'll see it. You know, it does some research. or write some code.β
βor use operator. You'll see this is the goal that the AI companies want. Systems that autonomously carry out your will.β
βagents bang on Blender, agents bang on Photoshop. And the reason for that is because those tools are first of all do a very good job.β
βif you haven't played with Manis or for most people who've listening, just take 03 and ask it a bunch of hard questions and open up its thinkingβ
βif you configure AI to work like a tutor and give it people classroom support in the right kind of way you get fairly big boost to learning outcomes.β
βif you're doing differential diagnosis you 100% want as a doctor you 100% want to be using AI to do this right like if you're not like it's already better than humans across most of that fieldβ
βLabs which takes you 10 to 15 minutes... can literally build you websites, dashboards, analytical charts, diagrams, study guides, slide decks, presentations.β
βThere's this thing in perplexity called tasks that you can just periodically run for yourself... I have a task that tells me the max seven stocks movement every day.β
βwe have this Dell portal on hugging face and we have all these open models and we qualified them on every kind of machine we haveβ
βGoogle has these Gemma models, GMMA, and they work really, really well on small machines.β
βHow many people have set up OpenClaw? Raise your hand. Oh my lord. That's about what a 20% of the audience here.β
βthe Atari learning environment is uh you know, it's born in Alberta and it underpins so much of what happened in dynamic reinforcement learning.β
βeventually migrating towards spinning up cloud instances using Visual Studio Code on remote systems and kind of working there.β
βthey're called April tags, these little fidials that robotics people use a lot uh to identify things...We could identify those and then we can pull out a nice rectified image of the screen.β
βI'm very bullish on computer use as a technology... I'm really hoping and I'm sort of excited about computer use technology actually maturing.β
βThere's open AI operator browser base is more like infrastructure but but it's interesting.β
βThere's another one called pig.dev for Windows machines.β
βyou could ask your computer, is there a proof of the remon hypothesis...in a machine checkable format like, coaul or lean or some formal verification language like thatβ
βLiveKit on the infra side.β
βif you go to the humanmovement.org there will be, you know, action groups and things that people can do there for actually taking action on this that are not just passively sharing news links, but likβ¦β
β5 hours later, I had 5 [music] hours of footage. I uploaded it to Da Vinci Resolve, which I use for video editing, and it [music] transcribed 3 GB of video into 317 kilobytes of my inner life.β
βHe'd redirected it to something which is folding at home, which is I think if you don't know it, you should look it up, but it's how you can allocate compute to helping solve health problems.β
βPerplexity, which is actually got a really great co-work competitor called Computer I've been playing with... The model council is like the greatest feature.β
βThe model council is like the greatest feature. You ask it a question, it will go to all three different major models and you can pick which ones including open source. Then it tells you where they diβ¦β
βWe put it into auto research and we doubled the click-through rate. And I was like, to Freeberg's point, this was many man months, tens of people.β
βGoogle has announced Google workspace studio to do AI automation and it's online and people are playing with it already. So they have joined the open claw party.β
βyou have a multiplatform approach where you build it in uh something like cotlin compos multiplatform um you write most of your code in thatβ
βLoad a chunk of [Ultachat_200k](https://huggingface.co/datasets/HuggingFaceH4/ultrachat_200k)β
β2. We embed the text with Gemini 2 EmbeddingβοΈβ
βI'm very excited about the live API that we've been working on for a while...we released it back in December...the model can you can talk to the models. The models can see what's on your screen.β
βHinton showed us a test he devised for Chat GPT4, the chatbot from a company called Open AI... You believe that Chat GPT4 understands? I believe it definitely understands. Yes.β
βWhen Anthropic came out with Sonnet Claude 3.5 that was a fundamental shift in the industry where the models got a lot better at coding...it was a start of vibe coding where you can put in a prompt anβ¦β
βSo far we're very impressed...it is good enough and the economics are so good that you can expend more tokens to get more intelligence.β
βwe have some of the best ones at deep mind we have Majoko right which we work withβ
βIf you are on Mac, I personally use Whisper Flow, which is $15 a month and adapts to whatever I'm writing.β
βYou got things like Aztec that are spinning up. They're going to be in mainet, but that's going to be its own separate rollup.β
βthe distributed history ideas from Portal like those are actually making their way into production.β
βThey benefit from work being done by people like Succinct. The whole ZKVM effort, you know, like Risk Zero, all of the other different ZK providers.β
βThe whole ZKVM effort, you know, like Risk Zero, all of the other different ZK providers.β
βI will open my Zoom app first to be able to record Zoom clips, which allows me to record my screen and talk and also transcribe my speech automatically.β
βIf you're ready to launch, download the 10 zero to ship VIP code prompts for free. Link in the description.β
βHow can I learn to be better at prompting? We have a YouTube channel.β
βI think the browser companies are interesting. So there's Strawberry, here's.com, there's DIA and Perplexity. Now I'm I'm very excited to see what happens to other companies.β
βWe recommend use of these cleaned CMB maps for low multipole (\ell < 30) studies.β
βWe present a new $\ell<30$ likelihood for use with the Cobaya package.β
βInteractive assets & ROI calculators: Teams create new calculators, demos, and interactive concepts in minutes, not weeks.β
βjust drop in a prompt or screenshot, and get a polished landing page right away.β
βIBM's Granite 4.0 is now on Replicateβ
βYour fabric uh which you've done you did some time ago actually but is still really really good.β
βAnd by the end you can either build the second brain yourself in notion or use [music] our template that sets it up in just 5 minutes.β
βI can speak for behavior and I think β Jim, you should speak for group. β for behavior like I explained earlier that one of the most important first step we took is to really embody the belief that AIβ¦β
βSo Rogo, which is a startup in the financial services space.β
βAnd the one I recommend is the save to notion web clipper which is really the easiest um to set up and operate.β
βThere's another startup called Accordance that's doing this in the tax space.β
βSo including optimism, arbitrum, base, tao, scroll, all of those, we'll be able to include again a a significantly larger amount of data.β
βSo including optimism, arbitrum, base, tao, scroll, all of those, we'll be able to include again a a significantly larger amount of data.β
βSo including optimism, arbitrum, base, tao, scroll, all of those, we'll be able to include again a a significantly larger amount of data.β
βSo including optimism, arbitrum, base, tao, scroll, all of those, we'll be able to include again a a significantly larger amount of data.β
βand also it works really well with this other tool ZKVMs right so zkVMs uh this is a technology that I think we've been talking about thinking about working on for basically the past like 10 yearsβ
βAnd there's a number of protocols that are working on this. So there is ERC7683.β
βAlso as there's starting like EIL is like another one that that a team at the EF is working on.β
βNow how do we do this? Basically we have to move away from optimistic roll-ups and we have to move toward ZK roll-ups.β
βBut today with Ziki snarks we do have the technology to provide it.β
βAnd on the right we have Aragon and Aragon is a DAO project like it's a a project that's writing software and tooling for Dows that has existed for a long time and they have been recently getting β¦β
βAnd by the way, everything I'm showing you here is included in our second brain notion template. If you want this exact setup, link is in the description.β
βSo there's also L2s, Aztec and IntMax I think are both very impressive.β
βbillions of people now interact with Gemini models, whether that's through AI overview, AI mode, or the Gemini app.β
βAnd this requires in protocol account abstraction. So things like EIP7701 and also Fossil which is a proposal that basically means that you have about 16 different actors in parallel that are able tβ¦β
βAnd this requires in protocol account abstraction. So things like EIP7701 and also Fossil which is a proposal that basically means that you have about 16 different actors in parallel that are able tβ¦β
βI remember having to buy books on Adobe Photoshop as a kid and then you'd read them to learn how to remove something from a from an image and how to fill it in and feather and all this stuff. Now anyoβ¦β
βAnd uh like the the like Nightfall and all of the efforts from EY have been I think excellent projects and you know they uh are just one example to show how much uh how much interest there is and evenβ¦β
βI mean either privacy pools type ideas or like ways of zero knowledge proving specific claims about yourself without revealing everything.β
βI mean I'd be curious to answer for other people like why don't you have 10 times more money in a real gunpool?β
βI mean inmax is uh I think like the the value ad there is uh actually yeah well in that case it is payments focused but it's privacy on top of the extreme scalability that's uh enabled by their versioβ¦β
βI mean uh also for like the wallet side for example right so I think you know the recent announcement started talking about kohaku and then also one of one but then the goal there is like not you knowβ¦β
βCustom internal tools: Teams spin up custom dashboards & admin tools, with AI features, to streamline any workflow.β
βI actually mean I mean I feel like actually Signal has been um gained a huge amount of like market share in terms of conversations within the crypto space over the course over the past year or so, rigβ¦β
βthe first launch it went viral because it was the first agent in the market that you could tryβ
βWith @Vanguard_Group, we tested sampling-based VQAs for portfolio construction.β
βHowever, there are some very clever ways called the quadratic civ and the number field civ that you can factor numbers on a digital computer much faster than trial division.β
βHowever, there are some very clever ways called the quadratic civ and the number field civ that you can factor numbers on a digital computer much faster than trial division.β