Sambhav Kothari @_sambhavkothari

Head of AI Productivity @bloomberg. Opinions are my own. https://t.co/2B0HQdJeqq linkedin.com/in/sambhav-kot… London, England Joined November 2016

Tweets

3K
Followers

394
Following

349
Likes

5K

Pydantic @pydantic

2 days ago

Pydantic AI version 1.0.0 is out! 🎉 github.com/pydantic/pydan…

6 53 244 14K 33

OpenAI defined AGI in 5 levels: with the recent "Reasoning" breakthrough, we're effectively past level 2 level-3 Agents are scaling quickly across labs, and 2025 is widely called the year of agents level-4 Innovators should begin advancing by late 2026

50 49 364 26K 102

Download Image

Aurimas Griciūnas @Aurimas_Gr

3 days ago

𝗔𝗜 𝗢𝗯𝘀𝗲𝗿𝘃𝗮𝗯𝗶𝗹𝗶𝘁𝘆 is now a must have in your tool belt as an AI Engineer. 𝗧𝗿𝗮𝗰𝗶𝗻𝗴 sits at the core of it, why is it important? Tracing and instrumentation of software have been around for decades now. With AI systems resembling regular software even more, we…

9 58 331 21K 368

Download Image

Haider. @slow_developer

2 days ago

one key point about the Gemini 3.0 release: demis says, Gemini version numbers match the pre-trained base they're built on Gemini 1.0, 2.0, and 3.0 are different base models, while Gemini 2.5 was a fine-tune on 2.0, so it needed less compute and shipped faster. he also said a…

19 6 305 21K 39

Aadit Sheth @aaditsh

3 days ago

This is literally everything you need to go from beginner to pro in AI

95 517 6K 470K 8K

Download Image

Google AI Studio @GoogleAIStudio

2 days ago

.@NanoBanana create an isometric diagram of google's TPU factory

147 150 4K 432K 632

prashant @prashantmital

2 days ago

myth #1: it's not possible to do some things with responses nope. responses is a superset of completions. anything you can do with completions, you can do with responses – plus more. for example, you can manage the conversation state manually (like in completions) OR let the…

7 5 150 37K 71

Download Image

prashant @prashantmital

2 days ago

feeling frustrated (and a little guilty): there’s still way too much confusion about @OpenAI's Responses API. this is partly on us: we haven’t always been clear about why we built it, how to use it, and why it matters. here's my attempt at setting the record straight. 👇

88 102 2K 525K 2K

prinz @deredleritt3r

2 days ago

Financial Times: OpenAI has co-designed its own chip with Broadcom. Chips are expected to ship next year. A mystery customer committed to $10B in chip orders from Broadcom earlier today. We now know who that is.

8 14 189 10K 28

Download Image

Lakshya A Agrawal @LakshyAAAgrawal

3 days ago

@Keivansamani AI using DSPy to solve tasks is amazing! Do checkout github.com/gepa-ai/gepa/b… and github.com/gepa-ai/gepa/b… btw, uses gepa to create a good DSPy agent generating both the prompts and architecture, improving score on MATH from 67% to 93%!

4 8 154 4K 199

Keivan @Keivansamani

3 days ago

Claude in cursor has realised it can construct DSPy agents to fulfil it's requests and has been doing that for the past couple of hours...

10 22 488 49K 294

Deedy @deedydas

3 days ago

Jane Street brought in $10B of revenue this Q with 3000 employees. That's $13.5M/yr of revenue per employee. Only OnlyFans ($37M), Valve ($19M) and Craigslist ($14M) are higher.

87 97 2K 196K 568

Download Image

Max Zeff @ZeffMax

2 days ago

OpenAI plans to launch an AI-powered hiring platform by mid 2026, putting the outfit in close competition with LinkedIn. The company also wants to start certifying people for "AI fluency." The ambition of this company is something else.

80 53 987 326K 241

Download Image

Darren Shepherd @ibuildthecloud

3 days ago

I did not see this coming. Create an AI pitch, exit before you can deliver. Well played.

The Browser Company @browsercompany

3 days ago

I did not see this coming. Create an AI pitch, exit before you can deliver. Well played.

444 265 3K 4.2M 630

Download Video

7 6 105 8K 2

jason liu @jxnlco

3 days ago

Don’t overload your coding agents with RAG. What we learned from Nik Pash @pashmerepat, head of AI @cline: 1) Modern models don't need embedding search 2) Agentic exploration works better 3) Reading full files beats chunking 4) Even Cloud Code and others abandoned RAG…

12 28 265 28K 350

Download Image

Chubby♨️ @kimmonismus

3 days ago

German "Who Wants to Be a Millionaire" Benchmark w/ Leading Models First place: Grok-4 75,6% Success Rate Second Place: GPT-5 71,1% Success Rate via lolcallama

29 32 452 50K 93

Download Image

Chubby♨️ @kimmonismus

3 days ago

It's coming. More and more reports about the coming unemployment are being published. "The Gerald Huff Fund for Humanity today released Impact of AI on Workers in the United States, a groundbreaking analysis revealing that artificial intelligence is on track to disrupt roughly…

35 83 352 28K 143

Download Image

swyx 🇸🇬 @swyx

2 days ago

Claude Code: no evals [well known code agent company]: no evals [well known code agent company 2]: kinda halfassed evals [leading vibe coding company]: no evals [ceo of company selling you evals]: mmmmm yess all my top customers do evals, you should do evals [vc's in love…

Lenny Rachitsky @lennysan

2 days ago

34 48 622 382K 890

115 73 1K 424K 1K

Download Image

Tibo @thsottiaux

4 days ago

Pro tip: Enable codex code reviews, it's been described multiple times to me as superhuman in its ability to spot subtle flaws. Navigate to chatgpt.com/codex, speedrun the onboarding, enable for your repos in the settings. Enjoy deep reviews on every PR.

83 179 1K 292K 982

Download Image

Pietro Schirano @skirano

3 days ago

I was so frustrated with the experience using nano-banana that I built a self-correcting image agent. The agent generates images and evaluates them against the original request until the goal is met. Like in this video, it was able to achieve the goal in 6 loops.