←
Michael344 posts
Michael avatar

Michael

@MichaelLee04

building! fellow @thielfellowship, prev startups, cs@stanford

◷ Joined March 2021▧ 20 Photos and videos
Michael avatar
Michael@MichaelLee04·Nov 17

First impression: 6.2 Astra and Sol are...unbelievably annoying? They constantly ask for approval for completely arbitrary and useless things, in a way I've never seen a model do. Some of the annoying exchanges I've had so far: > "How many intermediate files should I create for this RANDOM part of the refactor??" > "Waiting on your call now for which kind of format you want, between IncomprehensibleA and IncomprehensibleB" > "I know you already approved editing this, but just checking again that you want this?" wtf??

Carson Poole avatar
⟳ Michael retweeted
Carson Poole@CarsonPoole·Nov 10

Of course that’s your contention. You’re a first-year inference engineer. You just read SemiAnalysis’s blog and saw megakernels coming to TPUs and AMD, so naturally you believe that’s the way to get to speed of light. Next month, when you get your hands on an inference engineering cookbook—Baseten’s, probably—you’ll be singing from the treetops about the 500 nanoseconds of critical-path latency you save on every GEMM pipeline drain. That’ll last until sometime next year, when you watch a pirated geohotz lecture livestream and become convinced that PDL gives you a strict superset of megakernel capabilities. “Well, as a matter of fact, I won’t, because PDL introduces—” “PDL introduces unnecessary stalls because its synchronization mechanism is very coarse.” Yeah, I read that too. HazyResearch’s blog, “Look Ma, No Bubbles!” Just below Figure 2. Do you have any thoughts of your own on the subject, or should I just point my agent at the SGLang repo and have it do it for me?

Post attachment
Michael avatar
Michael@MichaelLee04·Nov 6

This is one of the worst pitches I've seen in my life. Complete audience / presentation mismatch that lost them in the first 5 minutes. Playing the space video and talking to these people about living on Mars as an intro was an insanely bad choice. These people are scared that they will be taken advantage of and hoodwinked by a slick corporate strategy with fast talking people who think they know better than them. That they will end up without significant benefits, with a noisy, pollutive, ugly datacenter that destroys their community, way of life, and property value. They've seen lots of negative media about AI companies, and have entrenched beliefs that are hard to overcome. They might be wrong on many of their beliefs, but that's the point! You need to change people's minds to make big change happen in their community. It needs to be clear that this is a win for them! No change is preferable to unknown change for most people. Making life multi-planetary is an inspiring story for a young engineer/product manager listening to Elon who wants to grow their career and equity value. But it might be quite literally the worst story to lead with for these community members. Guy presenting speaks quietly through an introduction, and then raises his voice to say he's the head of energy and supercomputing, and says multiple times that he's inspired by the SpaceX mission. They should have sent someone more prepared. This is politics, and they should have learned from political gatherings before them. What they needed to do: 1. Have a few trusted community leaders introduce them and say "Hey listen guys, we were skeptical at first, but we've really become convinced this would be good for us. It seems like there's a lot of benefits to this, they've addressed many of the downsides, and we should at least hear them out. This is Riley, and he works at SpaceX and he's here to answer your questions" 2. Cut out all of the multiplanetary shit and and inspiring videos. It should start with "Hi, SpaceX wants to build a datacenter near here, and we want your support to do it. We know you have a lot of questions, and we've actually gathered, over the past few days, the most common ones for this short presentation, and we'll take and answer more questions as we go through. But first, let's talk about why you should want this.." 3. SIMPLE. GO POINT BY POINT. Here is benefit #1, any questions? Here is Misconception #1, it's actually this. Any questions on this one? Am I saying this is an easy audience? Absolutely not. I'm not even saying he would have succeeded if he followed the above. But it would have been a hell of a better shot than whatever the fuck this was.

Please watch this video and ask yourself, who feels the need to silence respectful and open discourse?

21:58
Michael avatar
Michael@MichaelLee04·Nov 5

This is unbelievable - a must try

Two months ago, we introduced Atlas, an autoregressive world model with pixel-perfect camera control. Today, we're releasing Frameboy, a demo allowing you to explore worlds in real-time with a highly optimized version of Atlas, streaming from cloud GPUs at up to 15 fps.

0:46
Michael avatar
Michael@MichaelLee04·Nov 4

new admin, back to old zuck

0:05
Michael avatar
Michael@MichaelLee04·Nov 4

Mandate of heaven is with @t3dotcodes

I am very excited to share that Conductor is joining Meta! We have some big, crazy plans for the next interface. If you liked Conductor, you’re gonna love what’s coming. We’re super excited to work with @natfriedman and @alexandr_wang at MSL to build it. They’ve been on an absolute tear, and I’m convinced MSL has the highest talent density of all the labs. A huge thank you to our earliest believers @aaron_epstein, @ilyasu, and @nabeel. And to everyone here who tried Conductor, sent us feedback, and prompt requests - you made this so much fun. Thank you ❤️🚂 Choo choo!

0:06
Michael avatar
Michael@MichaelLee04·Oct 29

Someone please make a simple inference platform for LoRA-tuned models. Bill me by the token, you can be 150% the price of the base model. <1s additional latency needed. Load adapters from NVMe not in RAM so you can load tons of adapters from the same disk.

Michael avatar
Michael@MichaelLee04·Oct 28

Incredible. Google waited for everyone with taste to call AAV5 the benchmark most aligned with the performance they were seeing from the frontier. Then they dropped #1. I often hear rumors about model releases about a week or more in advance. This one took me by surprise!

Introducing Gemini 4 Carbon, it's a big jump over Argon which we released just 2 weeks ago. Gemini 4 Carbon is SOTA on cyber, long horizon tasks, ranks #1 on Artificial Analysis V5, significantly improves on usability in domains like coding, and is more aligned.

Post attachment
Greg Brockman avatar
⟳ Michael retweeted
Greg Brockman@gdb·Oct 26

Announcing three major scientific discoveries, one each in mathematics (harmonic analysis), computer science (complexity theory), and physics (quantum field theory). AI starting to help us improve our understanding of the universe! openai.com/index/building-sc…

Michael avatar
Michael@MichaelLee04·Oct 19

This should be the first diffusion model you try internally! Finally a diffusion model that will be near the Pareto frontier on price/performance, at fantastic speed. Lots of low latency applications for this one. I personally had doubts about scaling diffusion so very happy to be wrong about this. Would love to hear from people using this where they’ve found deficiencies.

1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec.

0:30
Michael avatar
Michael@MichaelLee04·Oct 17

I tried this new vision model yesterday on a few hundred prompts. This is not cheap at all, do not be fooled by the input and output price. It spends 20-45x the amount of tokens as models like Gemini Flash 3.8. It is also extremely slow. Like p50 was like 40s. Fastest request was 21s, with a trivial question. This is on the minimum low reasoning. It does really well and is a good model if you care only about accuracy! But it is not good/fast/cheap. It is good/slow/expensive.

Spent years watching models ace a single image and completely whiff on anything that moves or changes angle. Today we’re fixing that. Try Glimpse now, our first visual thinking model with competitive frontier performance compared to Astra on key public visual reasoning benchmarks.

1:30
Michael avatar
Michael@MichaelLee04·Sep 24

I feel attacked

We hired a new VP named Michael Lee. Unfortunately we already have a Michael Lee. He works in the warehouse. He's been here 11 years. New VP gets: michael.lee2@company.com He immediately messages IT. “Can you move the other Michael to michael.lee2 and give me michael.lee?” No. He asks HR. HR asks me if we can “prioritize seniority.” I explain that the warehouse Michael has eleven years of email history tied to that address and changing it because someone with a nicer title arrived would be insane. VP reluctantly keeps the 2. Warehouse Michael finds out. Changes his Teams display name to: Michael Lee (Original) By Friday the VP is being referred to as Sequel Mike. He has asked IT to intervene. Unfortunately nickname management is outside our support scope.

Michael avatar
Michael@MichaelLee04·Sep 21

I just opened LinkedIn for the first time in ~20 days. The feed is a Bacchanalian slop apparatus. It is painful to scroll through. There is neither pleasure nor valuable information, and it feels alien and weird. I go to LI now only for a quick targeted person search.

Michael avatar
Michael@MichaelLee04·Sep 15

I got access to Jev earlier today (thank you @hackgoofer). I have run ~5,000 requests so far, (which cost me around $2!), across classification, model routing, intent, steering, and many other things. tl;dr, Jev enables a new intelligent decision-making primitive, separate from deterministic code and LLM calls. This allows a class of decision-making that was neither suited to dumb, unintelligent code, nor to slow, expensive LLMs. It is super fast and cheap, and I think I will likely end up making a few Jev calls to every LLM call I make in my product. I think probably any company using LLM requests today can probably add a Jev call pre and/or post LLM calls to quite literally make their product much better for free, and have better tool calling behavior in many cases. I happened to have a personal benchmark for this as I’d been working on a ton of proactivity and classification tasks. I have been using the deepseek flash and more recently gpt 5.6 luna family of models as reasonably smart classifiers with low latency. Think questions like: - Did this conversation output contradict something they’ve mentioned before? - Should we send a followup message to this user based on our rules? - It’s been a few seconds of silence. Should we proactively send a message? In the past, I’ve been forced to write a bunch of what I call decision chains, mostly because an LLM classification call is very expensive in TIME (avg 4s), and less importantly can cost quite a bit if run on every message. Imagine a normal chat app. If you added 4s to every response to figure out if the response is good before sending it out, that ends up being pretty bad. So instead, I usually have to write some code that is a crude heuristic that runs quickly and decides whether to run the classifier. Obviously this sucks because you call the classifier many times that you don’t want to, which makes your p95 bad, and you also miss cases with the heuristic, and you also have to manage all of these weird chains. With Jev, it’s cheap enough, and fast enough (p50 ~150ms, p95 ~350ms in my testing!) that you can easily run it every turn. Heck you can reasonably run it before generation AND post generation, for any application that isn’t realtime voice, and still feel snappy. But this is just one use case. Think: smarter model routing, better context packing, better responses, better observability for intent/tags/safety, smarter retries and so much more. By simply thinking about the inputs and outcomes you want to enforce, you can use Jev to supercharge most model calls and reduce bad user outcomes. The more “quirks” a model has, the more valuable it ends up being. It’s a bit weird and unintuitive using Jev. Generally, you want to decrease the # of questions you ask a classifier, or it makes more mistakes. In fact, you might want to ask your questions kind of in a compound way, because the reasoning happens in a shared scratchpad of sorts. Adding questions muddies the scratchpad and makes it take longer. With Jev, you feel incentivized to go the other way, to formulate your query as a set of independent questions. It doesn’t feel like adding more questions decreases your performance on others. You can go a bit deeper to improve tool calls. Many model tools are things like turning on settings, or other things. You can easily improve models that are not very good at tool calling with Jev, by simply figuring out when to run them. You can do a pre-LLM call to figure out when to unfurl different tool definitions, in order to make your main LLM run better, you could run a background task with Jev + another LLM to reduce tool and context burden on your main LLM, and free it to be responsive. I’ve only scratched the surface of my testing but very excited!

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution

2:56
Michael avatar
Michael@MichaelLee04·Sep 15

If what they say in the research report is true, this will instantly replace at least a quarter of ALL LLM calls, basically capturing all tokens currently used for decision-making and scoring. Insanely excited to see if they come close to the level of usefulness, speed, and reliability they claim here.

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution

2:56