Krish Dasgupta (@officialKrishD) header
Krish Dasgupta (@officialKrishD) profile picture

Forever Learner | Building Reinforcement Learning Systems | Healthcare | Robots and Brains | Graph ML for Health | Privacy and Decentralized AI

Joined June 2020
Photo by Krish Dasgupta (@officialKrishD)Photo by Krish Dasgupta (@officialKrishD)Photo by Krish Dasgupta (@officialKrishD)Photo by Krish Dasgupta (@officialKrishD)Photo by Krish Dasgupta (@officialKrishD)Photo by Krish Dasgupta (@officialKrishD)Photo by Krish Dasgupta (@officialKrishD)Photo by Krish Dasgupta (@officialKrishD)Photo by Krish Dasgupta (@officialKrishD)Photo by Krish Dasgupta (@officialKrishD)
Krish Dasgupta retweeted
Following the discovery that eight eMudhra intermediate certificates use weak keys, Cloudflare is deploying mitigations and releasing client updates. We explain the impact, our response, and what customers and developers should do next. cfl.re/487DNbG
40
374
62
2,193
245,810
Krish Dasgupta retweeted
We recommend Cisco users disable the Smart Install (SMI) auto-loading feature on all NSS and DIB network devices to mitigate exploitation by remote attackers. Have you taken this action? Read more: media.defense.gov/2019/Jul/1… #CybersecurityAwarenessMonth
47
745
84
2,901
164,415
Krish Dasgupta retweeted
arXiv has updated our policy on rate limiting for all submitters. This update was made to fairly distribute moderator time & support the arXiv community of staff, volunteers, readers & authors. Please read our announcement to learn more: blog.arxiv.org/2026/10/01/up…
78
494
240
1,723
763,647
Krish Dasgupta retweeted
This is one of the most refreshing posts. A post where the Gov of India @GoI_MeitY is talking about the AI startup @Adalat_AI backed by @ycombinator Good for the YC portfolio to have such good clients in its YC batch companies. @garrytan - see this. Very apt post !
.@Adalat_AI (YC F26) is now backed by @ycombinator Serving 11 states & 6,000+ courtrooms with speech recognition, case management & paperless workflows. Expanding across India and the Global South. #IndiaAI #JusticeTech @AshwiniVaishnaw | @JitinPrasada | @SecretaryMEITY
2
2
208
Krish Dasgupta retweeted
Finding clinical signals in large-scale health sensing data streams is quite challenging. We introduce a biomarker discovery framework that pairs hypothesis generation with adversarial validation to discover rigorous digital biomarkers alongside human experts. Read more: goo.gle/4gavroG
7
73
4
507
27,876
Krish Dasgupta retweeted
Wonder what the “E” stands for in E2B and E4B models? Look no further! Learn more about these “effective” parameters, Per-Layer Embeddings (PLE), and how they boost a model's power without increasing the actual parameters used during computation in this 2-minute overview.
14
91
8
724
38,439
Krish Dasgupta retweeted
Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale. Model weights: huggingface.co/moonshotai/Ki… Tech report: github.com/MoonshotAI/Kimi-K… Tech blog: kimi.com/blog/kimi-k3
1,539
7,162
2,352
45,885
14,756,730
Excited for our first general model Inkling -- open weights, 975B, natively multimodal (text, image, audio). Available on Tinker, HuggingFace and partners. It is yours to personalize and use openly. It is yours.
Today, we are introducing Inkling. Inkling reasons efficiently across text, image, and audio modalities. We are making the full weights available. thinkingmachines.ai/news/int… Available today for fine-tuning on Tinker. Play with it in the Inkling Playground. 🧵
66
141
18
2,872
241,353
Can LLMs keep learning new skills without updating their weights? Modern LLMs can already master & combine many skills. But teaching them new skills in a scalable way without catastrophic forgetting remains an open challenge @icmlconf we introduce a new approach: skill neologisms
6
46
324
21,495
Krish Dasgupta retweeted
Replying to @deliprao
Delip - for light pun - Knowledge Distillation And yes, some contexts do get lost in knowledge distillation 😄
1
4
1
59
3,060
Unpopular Take / Opinion : TLDR : The Dopamine Effect of Closed Source AI driven Code Generation . I was talking with two of my juniors who have been “aggressively using Claude Code” for everything ( Opus 4.5). They are good coders but have been relying on Claude Code for much now. So, whenever they see the 100 USD reaching its limit, they quickly bounce to use more till they achieve what the code needs to be done. One of them ended up paying 85 USD extra. I asked them - was it worth it ? They just wanted to build something quick and wanted to build something efficient using AI. The problem was not money. The problem was not the usecase. The problem was unable to resist and still use more money above the capped prices such that they can code simple things too. They even called AI for container scripts. ( ideally they should have routed to a local code LLM for that ) But the fact is - the urge / the uncontrollable need was a dopamine addiction . And I am seeing this in many young coders , young developers. The intelligence is slowly getting replaced by addiction. You may disagree now, but give a year or two- it will be the fact. Solution ? Best is to orchestrate intelligence driven model allocation , as AI in code is not going anywhere. It is already getting integrated as I see.
1
1
2
199
Krish Dasgupta retweeted
llama-server -hf ggml-org/Qwen3.6-27B-GGUF --spec-default
21
53
8
657
75,997
Krish Dasgupta retweeted
What does it take to run 3, 5, or even 10 concurrent instances of Gemma 4 locally? We've open-sourced a demo letting you run multiple models side-by-side on your hardware. Gemma 4 26B A4B easily runs 10+ concurrent requests on a MacBook Pro M4 Max at 18 tokens/sec per request.
97
418
91
5,005
915,721
Replying to @_arohan_
Dwarkesh is not an AGI party bro. He's been pretty diverse in how he seeks alpha from across a wide variety of industries. I enjoy the hell out of a lot of his podcasts. This podcast though, his questions and narrative are very AGI Party takes, because I think they're very low-rank. That's my opinion.
6
9
1
387
58,695
Krish Dasgupta retweeted
Big theme here, healthcare AI needs memory, relationships, and context preservation, not just smarter models. Graph agents pushing that vision forward. @officialKrishD #AppliedHealthcareAISummit
1
1
139
Krish Dasgupta retweeted
Judging by my tl there is a growing gap in understanding of AI capability. The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code. But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along. So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions. TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.
The degree to which you are awed by AI is perfectly correlated with how much you use AI to code.
1,263
2,703
717
22,757
7,643,819
Krish Dasgupta retweeted
There's a high chance you'll NOT get access to the most intelligent models. when we say open source must win it's not about running a small offline model on the train...
This quoted post is unavailable.
1
2
58
2,783
Krish Dasgupta retweeted
People underestimate the level of collaboration that needs to happen for a model such as Gemma 4 to land Before the launch, we worked with HF, VLLM, llama.cpp, Ollama, NVIDIA, Unsloth, Cactus, SGLang, Docker, CloudFlare, and so many others This ecosystem is amazing 🔥
53
80
5
1,645
62,891
📢 The UN Open Source Week 2026 website is live! 22-26 June | UN Headquarters Join us for one of the leading global forums for open source collaboration. Explore the program, speakers & event info. 🔗 Learn more: unopensource.org 🔗 Register: forms.cloud.microsoft/pages/…
21
39
2,494