Kunal Tangri (@TangriKunal) header
Kunal Tangri (@TangriKunal) profile picture

Founder @farsight_ai · Building deep agents for enterprise deliverables · @MIT NLP · Ex-@huggingface · Forbes 30u30 AI

Joined April 2021
Photo by Kunal Tangri (@TangriKunal)Photo by Kunal Tangri (@TangriKunal)Photo by Kunal Tangri (@TangriKunal)Photo by Kunal Tangri (@TangriKunal)Photo by Kunal Tangri (@TangriKunal)Photo by Kunal Tangri (@TangriKunal)Photo by Kunal Tangri (@TangriKunal)Photo by Kunal Tangri (@TangriKunal)Photo by Kunal Tangri (@TangriKunal)
"Capture your institutional knowledge" has meant the same thing for 30 years: index the documents, search over them. But a document is the output of judgment with the judgment removed. The judgment is in the diff. The markup between the first draft and the one that ships. And almost every firm throws it away. @saranormous named the untrainable territory. At @farsight_ai we're building the instrument that lives inside it. The System of Judgment.
58
17
33
205
215,544
Back in 2021-2022 at Hugging Face, most of the AI work I did with banks was classic NLP. Now a lot of it is agentic. The first question hasn’t changed: where does the model run, and who can see the data? Wrote about it with @huggingface
The AI question in financial services has moved from which model to use to where that model is allowed to run. New post with @TangriKunal and @huggingface on why firms want to run models on their own infrastructure, and how governance gets them approved to run there. Read it here: farsight-ai.com/news/open-mo…
1
3
13
292
Kunal Tangri retweeted
Every firm has an edge. The best build on it. Something new is coming from Farsight. Stay tuned...
1
4
2
18
921
Ask a bank what its AI will cost next year and who sets that price. Most can't answer. We're partnering with @OrnnExchange so banks can source GPU capacity, see pricing built from transacted deals, and run @Farsight_AI in their own environment.
2
13
359
An M&A team and a capital markets team at the same bank can work very differently. @DeutscheBank now runs @farsight_ai on-premise across IB and capital markets, so each team builds from its own past work, inside the bank's own infrastructure.
2
1
9
403
Knew it. The paranoia was mechanistically justified
when you make typos in your prompts, you should fix them for the sake of the J-space. E.g.
2
8
1,253
Kunal Tangri retweeted
___________ | we're | | actually | | hiring ! | | __________ | \ (•◡•) / \ / —— | | _| |_ Open Roles: - Cloud Infra Engineer - Product Manager - Senior Frontend Engineer - Senior Software Engineer - Software Engineer - Founding SDR - Strategy & Ops Associate
31
5
2
125
38,193
Claude Fable not enterprise usable - no providers offer zero data retention
1
2
3
21
9,195
The most interesting part of Anthropic's RSI piece is the open-ended problems curve. Up 50 points in six months to 76%. The steepest climb of any task category. Open-ended problems are the ones that need judgement and taste - one of the big remaining human advantages. It's also the one improving fastest.
1
1
3
1,131
Does anyone else fix every typo before sending messages to Claude/Codex? I'm always paranoid that a slightly off-distribution input quietly drags the output off too
2
133
Pls fix, Codex is doing this on all of my chats rn
2
1
692
Based on the launch of Sites, GPT 5.6 must finally solve front end + is imminent
‼️‼️huge week for openai this is easily going to be my favorite week of the year huge updates to codex and 5.6 is a very special model. solved front end and personality. considerably better than 5.5 at everything. they could easily have called this gpt 6 digital agi is less than 12 months away (i’ve had this confirmed by the smartest dude in the lab with the mandate)
244
The bubbling AI cost crisis might be the best thing that ever happened to @huggingface. As frontier pricing keeps climbing, the incentive to build on smaller, efficient, domain-specific open models gets stronger every week. That shift runs directly through the open-source ecosystem, and Hugging Face is the center of gravity there. I spent time at HF early on, advising enterprises on everything from pre-training to deployment. Even then, the hardest question was always "when does open source actually win." I think rising closed-source costs are the forcing function that finally answers it. Hugging Face might be the most underrated company in AI right now. The biggest opportunity ahead of them is turning this moment into the right business model. If they capitalize, the positioning is unmatched
1
3
173
Life has only gotten busier since AI code assists started working. But today, I tasted what I want the future to look like. Spent 20 minutes scoping something on codex, then stepped away to cook dinner while it implemented. Worked flawlessly and dinner tasted great. This is the future that was promised
3
156