Tanishq Mathew Abraham, Ph.D.
@iScienceLuvr
Followers
78K
Following
100K
Media
2K
Statuses
17K
CEO @SophontAI | PhD at 19 (2023) | Founder, ex CEO @MedARC_AI | ex Research Director Stability AI | Biomed. engineer @ 14 | TEDx talk➡https://t.co/xPxwKTq6Qb
Joined December 2011
Help my parents out lol.
1/n Our amazing son Tanishq @iScienceLuvr turns 22 on June 10! 🎉. At 22, most are just graduating college—but he’s already earned a PhD and launched his medical AI startup @SophontAI 👨⚕️🔬. We’re so proud—and now we’re stumped on how to celebrate….
10
0
87
Nostalgic for GPT-3 and its absolutely hilarious generated greentexts.
44
244
7K
Have you heard of Cleo?. Cleo was an account on Math Stack Exchange that was infamous for dropping the answer to the most difficult integrals with no explanation. often mere minutes after the question was asked!!. For years, no one knew who Cleo was, UNTIL NOW!
38
360
6K
DeepSeek CEO's thesis is that in AI, there isn't any moat in being closed-source, but rather in having a talented team that can keep innovating
110
694
5K
Wait GPT-4o can just one-shot stuff like this?! That's impressive.
80
140
5K
I will attempt to explain the basic idea of how diffusion models work!. in only 15 tweets! 😲. Let's get started ↓.
80
700
4K
Anyone who thinks DeepSeek just came out of nowhere should see this graph. For each model on this graph, weights, code, and detailed papers were released. This is a team with a strong track record and has been working hard for a while. They didn't come out of nowhere.
75
670
4K
Very excited to share the news that I successfully defended my PhD research today! 🥳🎉. After 4 years 8 months in the @UCDavisGrad @UCDavisBMEGG graduate program,. I am now Dr. Tanishq Mathew Abraham (at 19 years old)!!
261
219
4K
Awesome and surprising things you can do with Jupyter Notebooks ⬇.
47
605
3K
Okay so this is so far the most important paper in AI of the year
52
354
3K
I got to try GPT-4's multimodal capabilities and it's quite impressive! A quick thread of examples. Let's start out with solving a CAPTCHA, no big deal
67
483
3K
US has teams building frontier models (OpenAI, Anthropic, etc.). Europe has teams building frontier models (Mistral, DeepMind, etc.). China has teams building frontier models (DeepSeek, Alibaba, etc.). What about India?? Why is India so behind on building SOTA foundation models?.
330
171
3K
how does someone solve Advent of Code problem in 9 seconds??!!
137
60
3K
Are you wondering how large language models like ChatGPT and InstructGPT actually work?. One of the secret ingredients is RLHF - Reinforcement Learning from Human Feedback. Let's dive into how RLHF works in 8 tweets!.
40
523
3K
This is currently the most important network in deep learning!. From helping to power search for billions of users to better understanding proteins, it does it all! . Here are 10 of the best resources to help you learn about the attention mechanism & Transformer network ⬇⬇⬇
26
565
3K
Diffusion language models are SO FAST!!. A new startup, Inception Labs, has released Mercury Coder, "the first commercial-scale diffusion large language model". It's 5-10x faster than current gen LLMs, providing high-quality responses at low costs. And you can try it now!
68
266
3K
Language Models Use Trigonometry to Do Addition. "We first discover that numbers are represented in these LLMs as a generalized helix, which is strongly causally implicated for the tasks of addition and subtraction, and is also causally relevant for integer division,
58
357
2K
A new tutorial on RL by Kevin Patrick Murphy, a Research Scientist at Google DeepMind who also wrote several comprehensive, well-regarded textbooks on ML/DL. This ought to be a good read 👀
18
277
2K
Just discovered this AMAZING website - "Deep Learning Drizzle". It is a constantly-updated list of machine learning/deep learning course materials that are taught by domain experts and are available for FREE!. Check it out here →
21
678
2K
Diffusion Models Are Real-Time Game Engines. abs: project page: Google presents GameNGen, the first game engine powered entirely by a neural model that enables real-time interaction with a complex environment over long trajectories
89
408
2K
Training Large Language Models to Reason in a Continuous Latent Space. Introduces a new paradigm for LLM reasoning called Chain of Continuous Thought (COCONUT). Extremely simple change: instead of mapping between hidden states and language tokens using the LLM head and embedding
51
304
2K
"A Manga Guide to DeepSeek-V3 Technical Report". from now on this is how I will post all papers 🤣
45
186
2K
ICLR 2023 (a top ML/AI conference) submissions have been released, and do you know what that means? . Time for mind-blowing papers! 🤯↓.
37
357
2K
This YouTube video by @welchlabs is a very approachable, concise explanation of LLM mechanistic interpretability and sparse autoencoders (SAEs). I highly recommend checking it out!
18
227
2K
Google is releasing a diffusion language model let's goooooooo!
39
96
2K
Large Language Diffusion Models. Introduces LLaDA-8B, a large language diffusion model that pretrained on 2.3 trillion tokens using 0.13 million H800 GPU hours, followed by SFT on 4.5 million pairs. LLaDA 8B surpasses Llama-2 7B on nearly all 15 standard zero/few-shot learning
38
289
2K
Thinking about creating a medical AI group chat here on Twitter. If you're a researcher/engineer/clinician/etc. working in AI or medical AI, let me know if you're interested in joining!.
991
86
2K
I had a great time chatting with @karpathy yesterday! We discussed a range of topics, from synthetic biology to ML conferences. What's funny is that we are about 2 hrs from each other but we got to meet up in Amsterdam airport! 😄
23
43
2K
He is one of the best AI researchers of all time. A good reminder that you shouldn't over-index on signals like GitHub contributions.
20
60
2K
The inventors of flow matching have released a comprehensive guide going over the math & code of flow matching!. Also covers variants like non-Euclidean & discrete flow matching. A PyTorch library is also released with this guide!. This looks like a very good read! 🔥
7
287
2K
How the machine learning community feels after PaLM and DALL·E 2 during this week:
16
160
2K
A new YouTube video by Welch Labs that gives an awesome walk-through of how multi-latent attention works (one of the innovations of DeepSeek)!. Check it out!
12
207
2K
Research in AI is surprisingly more accessible to people with different backgrounds compared to other fields. Anyone (w/ relevant experience) can contribute to impactful research. Here are 5 research orgs you can join to contribute to real, open research in deep learning ↓.
32
324
2K
@12exyz tbf o3 isn't public and those benchmarks can't be verified yet so I think it's understandable that they haven't included it yet.
16
9
2K
The livestream demo is not the only cool part about GPT-4o. Remember, GPT-4o is an end-to-end trained multimodal model!. No one is reading the GPT-4o blog post which highlights so many other cool features. SEE MORE FEATURES GPT-4o HAS ↓.
16
147
2K
They're blaming it on a single rogue employee lol.
We want to update you on an incident that happened with our Grok response bot on X yesterday. What happened:.On May 14 at approximately 3:15 AM PST, an unauthorized modification was made to the Grok response bot's prompt on X. This change, which directed Grok to provide a.
90
29
2K
Course material for an MIT class "Introduction to Flow Matching and Diffusion Models", looks great if you want a principled and hands on understanding of diffusion models/flow matching
24
223
2K
People are discovering how autoregressive image generation works lol.
one of the most annoying grok experience . watching it slowly reveal image from top to down . just show the image as it's made raw !
32
24
2K
If someone says RL is so simple and easy, ask them if they've read these blog posts 😄
27
158
2K
So @StableDiffusion has various options and controls and one of the main ones is the sampler used for generation. Let's talk a little bit about these samplers since this has some interesting and unexpected effects on generated image quality (below image from subreddit)🧵
28
225
1K
Meta researchers used AI to predict the text a person was typing just from non-invasive brain recording!. With EEG, their "Brain2Qwerty" model gets 67% of the characters wrong, but magnetoencephalography (MEG) shows much better performance, instead only getting 32% of the
151
266
2K
Part of the reason the public is shocked is that most of them have only played with ChatGPT 4o (in the free plan) so when they try a reasoning model like R1 they think China has made this incredible leap in abilities over Americans. Public doesn't know about o1, o3, Claude, etc.
162
87
1K
Yesterday was my last day at Stability AI. I had a great time over the past 2.5 years working with amazing colleagues on the cutting edge of AI research and development but now it is time for new adventures. I first joined Stability as a 19-year-old while wrapping up my PhD,
92
41
1K
I appreciate DeepSeek providing examples of failure, especially since these are ideas that have been widely discussed for achieving o1-style models. This is very rare to see in AI papers.
20
157
1K
Google coming after OpenAI, Meta, Apple, pretty much everyone today.
29
60
1K
The Tesla team discussed how they are using AI to crack Full Self Driving (FSD) at their Tesla AI Day event. They introduced many cool things:.- HydraNets.- Dojo Processing Units.- Tesla bots.- So much more. Here's a quick summary 🧵:.
12
243
1K
After you train a machine learning model, the BEST way to showcase it to the world is to make a demo for others to try your model!. Here is a quick thread🧵on two of the easiest ways to make a demo for your machine learning model:.
14
227
1K
Microsoft Phi-4 is announced!. It's a 14B parameter LM trained heavily on synthetic data, with very strong performance, even exceeding GPT-4o on GPQA and MATH benchmarks!. Currently available on Azure AI Foundry, will be on HuggingFace next week
23
191
1K
What matters most when training a neural network is how well it generalizes to unseen data. For neural networks, it turns out there's a simple principle that can allow you to understand model generalization. (1/18). A thread ↓.
19
251
1K
Vladimir confirmed to an investigator he indeed was Cleo, providing more details about the whole saga. He was coming up with various hard integrals to solve and he had an idea of what the solutions would be and he wanted to confirm his solution was correct, so this is how he.
13
21
1K
Are you wondering how the new Mamba language model works?. Mamba is based on state-space models (SSMs), a new competitor to the Transformer architecture. Here are 5 resources to help you learn about SSMs & Mamba! ↓↓↓.
23
189
1K
The devil works fast but lucidrains works faster
16
76
1K
I have EXCITING news:. I've started a company!. Introducing Sophont. We’re building open multimodal foundation models for the future of healthcare. We need a DeepSeek for medical AI, and @SophontAI will be that company!. Check out our website & blog post for more info (link
113
113
1K
Some people need to stop being so math-brained when writing their papers. Using 10 pages of math to explain 10 lines of code is super annoying.
53
39
1K
YOLO v12???? you've got to be kidding me, why is there a new YOLO model like every 6 months lol
52
65
1K
Spotted at #NeurIPS2023: Disney Research demos a RL-trained remote control robot! Super impressive work!
29
148
1K
The @PyTorch team is developing a library for large model training called torchtitan 👀. They have scripts to train Llama-3 from scratch. The library went public today on GitHub but it is still in pre-release state & active development. Check it out →
6
199
1K
xAI releases Grok-1. blog: code: Base model trained on a large amount of text data, not fine-tuned for any particular task. 314B parameter Mixture-of-Experts model with 25% of the weights active on a given token. Trained from
42
264
1K
Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach. We study a novel language model architecture that is capable of scaling test-time computation by implicitly reasoning in latent space. Our model works by iterating a recurrent block, thereby unrolling
36
188
1K
GPT-4 release.Med-PaLM2 announcement.PaLM API release.Claude API release
8
143
1K
One of my favorite YouTubers, @3blue1brown, has put out an incredible video explainer about the attention mechanism! . I highly recommend checking it out!
9
131
1K
Annotated PyTorch Paper Implementations by @labmlai is an AMAZING resource:. • Deep learning papers explained in-depth with code side-by-side.• Constantly updated with some of the latest papers!.• 100% free and open-source! . Check it out here →
4
226
1K
[MASK] is All You Need. New paper from CompVis group, introduces a new method called Discrete Interpolants that builds on top of discrete flow matching, and connects it to masked autoregressive models. Achieves SOTA performance on MS-COCO, competitive results on ImageNet 256, and
6
129
1K
Google announces Med-Gemini, a family of Gemini models fine-tuned for medical tasks! 🔬. Achieves SOTA on 10 of the 14 benchmarks, spanning text, multimodal & long-context applications. Surpasses GPT-4 on all benchmarks!. This paper is super exciting, let's dive in ↓
20
209
1K
How much do language models memorize?. "We formally separate memorization into two components: unintended memorization, the information a model contains about a specific dataset, and generalization, the information a model contains about the true data-generation process. When we
8
175
1K
born just in time to build ChatGPT wrappers.
born too late to develop transformers.born too early to develop AGI.
20
76
1K
Perspectives on the State and Future of Deep Learning -- 2023. abs: Leading AI researchers write about their thoughts on the future of deep learning
7
129
358
I would pay hundreds of dollars per month to cut down or eliminate the amount of sleep my body requires. more people need to be working on this fr.
210
45
1K
I am telling you guys. if you really want to truly grasp diffusion models. you MUST read all of @sedielem's blog posts!!!
14
115
1K
xAI: Announcing Grok. Grok-1 model card: The first model, Grok-0, was a 33B autoregressive LLM, and approached Llama-2-70b with half the training resources. Grok-1 surpasses that, achieving 63.2% on the HumanEval coding task and
22
136
907
Claude, @AnthropicAI's powerful ChatGPT alternative, was trained with "Constitutional AI". Constitutional AI is particularly interesting since it uses less human feedback than other methods, making it more scalable. Let's dive into how Constitutional AI works in 13 tweets!.
18
117
900
Given this model was trained on 100k H100s, this model better be wildly amazing. .
Grok 3 release with live demo on Monday night at 8pm PT. Smartest AI on Earth.
46
28
891
Nothing is revealed in this 10 min long video.
thrilled to be partnering with jony, imo the greatest designer in the world. excited to try to create a new generation of AI-powered computers.
81
3
893
What happened to Julia being the future of ML? Why didn't it succeed?.
128
23
878
Transformers Can Do Arithmetic with the Right Embeddings. abs: code: Improves Transformer's arithmetic abilities by adding an embedding to each digit that encodes its position relative to the start of the number. "We find that
14
140
858
so you guys know RL can be used for more than just math and coding, right?.
56
35
857
The Stable Diffusion 3 paper is here 🥳. I think my colleagues have done a great job with this paper so thought I'd do a quick walk-thru thread (1/13)↓.
Scaling Rectified Flow Transformers for High-Resolution Image Synthesis. paper: blog post: The Stable Diffusion 3 paper is here! . Introduces a novel diffusion transformer arch and uses the rectified flow formulation, scales up to
15
124
837
Perception Encoder: The best visual embeddings are not at the output of the network. "we find that contrastive vision-language training alone can produce strong, general embeddings for all of these downstream tasks. There is only one caveat: these embeddings are hidden within the
8
138
834
Breaking news!. Alec Radford departs OpenAI!. As one of their star researchers, he was first author on GPT, GPT-2, CLIP, and Whisper papers.
25
80
808
Reflecting a bit, 2023 was an especially good year for me. Wins included:.• Completed my PhD at the age of 19 (@ucdavis).• Founded @MedARC_AI and growing it to 2 employees and a community of >2.5k members and counting!.• Joined @StabilityAI as a Research Director full-time.•
32
26
803
This is your neural network looking for a local minimum during training😂.#DeepLearning #AI @ai_memes
7
91
774
One of the accounts, Laila Podlesny, had an email address associated with it, and by trying to fake log into the Gmail and obtaining the backup recovery email, someone figured out that Vladimir Reshetnikov was in control of Laila Podlesny. Based on other ineractions from.
2
17
812
The Chinese "sucked the knowledge" from American AI 🤦♂️. I don't know what to say.
115
31
801
Perplexity used to have this incredible tool to search Twitter, I absolutely loved it. Unfortunately, they had to shut it down 😭. I am BEGGING @X, @xAI to please add this functionality directly into Twitter. It's probably pretty straightforward to implement with Grok.
Introducing Bird SQL, a Twitter search interface that is powered by Perplexity’s structured search engine. It uses OpenAI Codex to translate natural language into SQL, giving everyone the ability to navigate large datasets like Twitter.
37
46
791
The math looks scary, I remember being very intimidated by it too. but the underlying concepts are really simple. and when you understand it, you can appreciate how beautiful diffusion models really are.
27
46
802
So, I've heard people say anyone could have built ChatGPT. I think this is disingenuous. ChaGPT isn't just GPT-3 w/ a chat interface on top of it. The closest base model on the OpenAI API is probably text-davinci-003, but it was only released a day before ChatGPT! (1/9)
23
82
765
People noticed that the same few people were interacting with Cleo (asking the questions Cleo answered, commenting, etc.), a couple of them only active at the same time as Cleo as well. People were wondering maybe someone is controlling all these accounts as alts
3
13
796
Some people claim that Project Stargate must be a complete waste of money, look at what DeepSeek did with so little money!. This is such a bad take!. DeepSeek provided NO EVIDENCE that scaling is hitting a wall!! More compute for an R1 training approach would likely give better.
96
55
787
and then he wins a Nobel Prize the next day lol.
Massive congratulations to my good friend and former Google colleague @geoffreyhinton on winning the Nobel Prize in Physics (with John Hopfield)! Incredibly well deserved, Geoff laid the foundations for the deep learning revolution that underpins the modern AI field.
11
39
770
Thru Discord:.- I have gotten a job.- I have found amazing colleagues.- I have collabed w/ people to write high quality papers in prestigious AI conferences.- I have met many amazing friends. Twitter has a lot of value to me, but Discord completely changed my life.
22
21
766
Overall it's an interesting saga, and interesting investigation from some online sleuths, you should definitely check out the investigative video from @joeMakinYaCrazy:.
6
24
775