Apoorv Sharma

Entrepreneur | Ex ML Eng @Meta @Uber @Pinterest @Twitter | Built core HNSW update algo with Yury (most popular Vector search algo)

[Jeff Huber @jeffreyhuber]

I’ve been looking forward to today for 3 years

Today we’re announcing Foundation - Chroma’s solution to memory

Our research preview of this technology builds self-improving memory from your agent sessions. Try it out at https://t.co/DpynS0vnSd

[Apoorv Sharma @apoorv_nsit]

@vijayshekhar @xai https://t.co/JEedGcOrEW

[AI at Meta @AIatMeta]

We’re sharing the next major milestone in our non-invasive brain-to-text decoder research: Brain2Qwerty v2.

Building on v1, which was published today in @Nature, Brain2Qwerty v2 is the highest-performing end-to-end pipeline capable of real-time sentence decoding from raw brain signals. It advances beyond character-level performance to decoding words and semantics, enabling accuracy for overall communication.

We believe this research has the potential to make a real difference for the millions of people who suffer from brain lesions or disorders that prevent them from communicating.

🧵👇

[Apoorv Sharma @apoorv_nsit]

@stevenkplus1 Codex/Claude code is good, why to use devin?

[Apoorv Sharma @apoorv_nsit]

@theodormarcu But why?

[Apoorv Sharma @apoorv_nsit]

@OnlyTerp @cognition Seems like some marketing campaign is going on before funding.

[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

[Apoorv Sharma @apoorv_nsit]

@JoshKale How to short it?

[Apoorv Sharma @apoorv_nsit]

@zivdotcat Codex

[Apoorv Sharma @apoorv_nsit]

@ns123abc What will cognition/devin do? They are primarily in cobol migration https://t.co/RMv4dIC1yW

[Apoorv Sharma @apoorv_nsit]

@windsurf Will you train on this data?

[Apoorv Sharma @apoorv_nsit]

@cognition @cursor_ai @opencode @vercel @julesagent @AmpCode @Cloudflare @savarlamov @leerob Code is just the artifact, the 'trace' is the intelligence. We all know the real value here is capturing the trajectories Git throws away. Since robust synthetic data for training is so hard to generate, this becomes the gold standard dataset for the next era.

[Chroma @trychroma]

Introducing Chroma Cloud: an open-source serverless search database that is fast, cost-effective, scalable, and reliable.

[CurieTech AI @CurieTechAI]

#DataWeave benchmark reveals -- GPT-5 underperforms GPT-4.1 in one-shot tasks but excels in multi-step reasoning via agent mode.
Our fine-tuned #CurieTech AI model demonstrates superior performance over generic agents across real-world transformation scenarios #MuleSoft

[BURKOV @burkov]

If Apple buys Perplexity, Cook will not only have demonstrated that he's a poor product executive, but also that he's a poor businessman.

Perplexity is nothing but a cheap wrapper around a poor-quality generic search engine and a language model hyped by people who wouldn't feel it if they were chowing shit.

[Apoorv Sharma @apoorv_nsit]

@deedydas Distribution of users (geography, age, service- delivery vs mobility etc.) using waymo vs other services will be very different. This comparison may not be very useful

[Apoorv Sharma @apoorv_nsit]

@deedydas Codeium inline completions are not good. Augment is way better.

[Bindu Reddy @bindureddy]

Grok-3-Thinking Scores Way Below o3-mini-high For Coding on LiveBench AI

Grok-3 is a good model, and OpenAI bashers love Grok-3 thinking for obvious reasons. 😉

Objectively, however, it scores WAY BELOW o3-mini-high for coding, and it takes forever to answer the most basic coding questions.

o3-mini-high - 82.74
grok-3 thinking - 67.38

On the plus side, it's a bit above Sonnet, though Sonnet is like 10-100x FASTER.

TLDR: Grok-3 is good for great banter on X but isn't a great model for developers.

[Apoorv Sharma @apoorv_nsit]

@realSharonZhou @LaminiAI @huggingface Thank you. In an enterprise setting, how do you come up with how many facts to cluster data in. Ex: if I am building a model for sql - are clustering sql queries into facts and then training on them?

[Apoorv Sharma @apoorv_nsit]

@realSharonZhou @LaminiAI @huggingface How do you decide the number of experts and ways to craft data for each?

[Apoorv Sharma @apoorv_nsit]

@siddrrsh @AkshGarg03 @mustafaaljadery Plagiarism. Not cool 👎