rami

data scientist @Amazon
MEng @imperialcollege

[Omar Khattab @lateinteraction]

this is such an impressive result: search over ~600,000,000 colbert vectors in 10 milliseconds, with a *single* CPU core.

and since this algorithm has sub-linear latency, there’s no excuse for anyone up to tens of billions of tokens

[Citrini @citrini]

People keep confusing a bubble with “stocks go up and get overvalued”. A bubble is when when a prevailing trend and a prevailing misconception about that trend interact reflexively, each reinforcing the other until the gap between perception and reality becomes unsustainable.

A bubble is not when everyone realizes that right now every iota of AI demand eventually, at some point upstream, must move through memory OEMs. Nor is it when estimates continue rising because things are better than expected. And it’s not just when stocks trade expensive to historical valuations.

The reason behind the moves in the AI infrastructure layer so far have been simply that we don’t have enough. They’ve been driven by the fundamental reality more than the perception of the future. It’s why the bulk of the most bullish parts of this cycle have been lumpy and centered around earnings season when companies uniformly come out and confirm there’s still not enough. In the bubble, the reality is driven by the market - not the other way around.

Everyone keeps saying “people are gonna freak out if it’s not a bubble!”. I think that’s silly, we have a transformative new technology that needs crazy capital to fuel it coming to fruition, that has and always will result in a bubble as long as we have financial markets.

But if you want to call the top in a bubble, you need a much stronger view on what the misconception is and what negative catalyst forces broad perception to align with realizing it than you do on valuation.

[Citrini @citrini]

Strait of Hormuz: A CitriniResearch Field Trip

The Field Report from Analyst #3 is live.

https://t.co/rwJffd8l81

[Cursor @cursor_ai]

We're releasing a technical report describing how Composer 2 was trained.

[Chroma @trychroma]

Introducing Chroma Context-1, a 20B parameter search agent.

> pushes the pareto frontier of agentic search

> order of magnitude faster

> order of magnitude cheaper

> Apache 2.0, open-source

[Cursor @cursor_ai]

Semantic search significantly improves coding agent performance.

For very large codebases, Cursor's indexing process is now several orders of magnitude faster.

https://t.co/Q1cKl71ytU

[Charles Packer @charlespacker]

Continual learning in token space: https://t.co/WxThWP2Kpz

[rami @ralshuk]

@densancar Excited to see whats next

[Google Research @GoogleResearch]

Introducing Nested Learning: A new ML paradigm for continual learning that views models as nested optimization problems to enhance long context processing. Our proof-of-concept model, Hope, shows improved performance in language modeling. Learn more: https://t.co/8wvV9vyA5V

@GoogleAI

[Andrej Karpathy @karpathy]

Agency > Intelligence

I had this intuitively wrong for decades, I think due to a pervasive cultural veneration of intelligence, various entertainment/media, obsession with IQ etc. Agency is significantly more powerful and significantly more scarce. Are you hiring for agency? Are we educating for agency? Are you acting as if you had 10X agency?

Grok explanation is ~close:

“Agency, as a personality trait, refers to an individual's capacity to take initiative, make decisions, and exert control over their actions and environment. It’s about being proactive rather than reactive—someone with high agency doesn’t just let life happen to them; they shape it. Think of it as a blend of self-efficacy, determination, and a sense of ownership over one’s path.

People with strong agency tend to set goals and pursue them with confidence, even in the face of obstacles. They’re the type to say, “I’ll figure it out,” and then actually do it. On the flip side, someone low in agency might feel more like a passenger in their own life, waiting for external forces—like luck, other people, or circumstances—to dictate what happens next.

It’s not quite the same as assertiveness or ambition, though it can overlap. Agency is quieter, more internal—it’s the belief that you *can* act, paired with the will to follow through. Psychologists often tie it to concepts like locus of control: high-agency folks lean toward an internal locus, feeling they steer their fate, while low-agency folks might lean external, seeing life as something that happens *to* them.”

[Andrej Karpathy @karpathy]

Excited to release new repo: nanochat!
(it's among the most unhinged I've written).

Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single, dependency-minimal codebase. You boot up a cloud GPU box, run a single script and in as little as 4 hours later you can talk to your own LLM in a ChatGPT-like web UI.

It weighs ~8,000 lines of imo quite clean code to:

- Train the tokenizer using a new Rust implementation
- Pretrain a Transformer LLM on FineWeb, evaluate CORE score across a number of metrics
- Midtrain on user-assistant conversations from SmolTalk, multiple choice questions, tool use.
- SFT, evaluate the chat model on world knowledge multiple choice (ARC-E/C, MMLU), math (GSM8K), code (HumanEval)
- RL the model optionally on GSM8K with "GRPO"
- Efficient inference the model in an Engine with KV cache, simple prefill/decode, tool use (Python interpreter in a lightweight sandbox), talk to it over CLI or ChatGPT-like WebUI.
- Write a single markdown report card, summarizing and gamifying the whole thing.

Even for as low as ~$100 in cost (~4 hours on an 8XH100 node), you can train a little ChatGPT clone that you can kind of talk to, and which can write stories/poems, answer simple questions. About ~12 hours surpasses GPT-2 CORE metric. As you further scale up towards ~$1000 (~41.6 hours of training), it quickly becomes a lot more coherent and can solve simple math/code problems and take multiple choice tests. E.g. a depth 30 model trained for 24 hours (this is about equal to FLOPs of GPT-3 Small 125M and 1/1000th of GPT-3) gets into 40s on MMLU and 70s on ARC-Easy, 20s on GSM8K, etc.

My goal is to get the full "strong baseline" stack into one cohesive, minimal, readable, hackable, maximally forkable repo. nanochat will be the capstone project of LLM101n (which is still being developed). I think it also has potential to grow into a research harness, or a benchmark, similar to nanoGPT before it. It is by no means finished, tuned or optimized (actually I think there's likely quite a bit of low-hanging fruit), but I think it's at a place where the overall skeleton is ok enough that it can go up on GitHub where all the parts of it can be improved.

Link to repo and a detailed walkthrough of the nanochat speedrun is in the reply.

[Thinking Machines @thinkymachines]

LoRA makes fine-tuning more accessible, but it's unclear how it compares to full fine-tuning. We find that the performance often matches closely---more often than you might expect. In our latest Connectionism post, we share our experimental results and recommendations for LoRA.

https://t.co/fYV4FPi71m

[Mira Murati @miramurati]

Sharing our second Connectionism research post on Modular Manifolds, a mathematical approach to refining training at each layer of the neural network

[Infinite Books @infinitebooks]

Nietzsche, what a line

[Google DeepMind @GoogleDeepMind]

We’re announcing a major advance in the study of fluid dynamics with AI 💧 in a joint paper with researchers from @BrownUniversity, @nyuniversity and @Stanford.

[Naval @naval]

The only real test of intelligence is if you get what you want out of life.

[tobi lutke @tobi]

Both DSPy and (especially) GEPA are currently severely under hyped in the AI context engineering world

[Tenobrus @tenobrus]

i'm suddenly feeling my biological clock ticking insanely hard except instead of for having kids it's for having $1 mil liquid by 30

[Mayank @_mayankkishore]

99% of SF founders forget to read the last line of pg’s essay on How to Do Great Work

[Jash Dholani | link in bio has my books! @oldbooksguy]

This quote has been stuck in my head for hours