Tanishq Mathew Abraham, Ph.D.
21.1K posts
Tanishq Mathew Abraham, Ph.D.
@iScienceLuvr
CEO @SophontAI | Founder @MedARC_AI | PhD at 19 (2023) | ex Research Director Stability AI | Biomed. engineer @ 14 | TEDx talk➡https://t.co/xPxwKTq6Qb
Katılım Aralık 2011
1.5K Takip Edilen89.9K Takipçiler
hell yeah raised more money 💸🥰
the journey of lightsaber powerplant continues 🔆
English
@mihirneal No, due to a stupid policy on OpenAI's part, I cannot transfer my chats.
English
Going to lose access to both my ChatGPT and Claude accounts. (Due to stupid policies by both co's)
It's very sad, I've had both accounts since 2022!
Perhaps I'm being a bit dramatic but it feels like a part of me and my life history is going away :(
Time for a reset, I guess
English
@ChaseBlagden Because I created my account with my uni email I cannot transfer it to a different email even though the account is a Personal account and never connected to the uni otherwise. Apparently OpenAI lets unis claim accounts and restrict transfer with no warning to the user.
English
While AI gives me a lot of hope to find cures for complex chronic conditions like mine, this potential will only be realized if we get enough biological data to understand the mechanisms of these conditions. That’s what we’re trying to do at @ChronicleBioAI:
fortune.com/2026/08/03/fid…
English
My ResearchGate profile is still a picture of 12-yr-old me
English
Honestly we need all these AI art hype influencers to stop posting until an actually good masterpiece is created with AI.
I do think it's possible with that in the hands of people with actual creative taste, good AI art can be made. But AI hype influencers aren't that!!!
el.cine@EHuanglu
if a space odyssey was made in 2026 with AI
English
I literally cannot escape AI discourse anywhere I go, it's insane.
I was having dinner with my family at a restaurant in Aspen, Colorado. So this isn't a tech center like SF.
But the family seated at the table next to us is arguing about AI!
Brother and sister argument went something like
"I use AI for studying and homework all the time, it's a very useful tool"
"You are becoming too reliant on it you're destroying your mind and it's bad for the environment!"
"It's not worse than search which uses AI"
"Search doesn't use AI"
"Yes it does"
"You can turn it off"
"It's still AI"
"I don't want to argue with someone who's becoming so dependent on AI, it's pointless"
English
finding out hank green, the guy responsible for teaching me so much science and history and research whose whole thing was using credible sources, is now using chatgpt for "research" is so deeply interesting!!
✨bri✨🧡🤍🩷@bricheesy
finding out hank green, the guy responsible for teaching me so much science and history and research whose whole thing was using credible sources, is now using chatgpt for "research" is so deeply saddening
English
A good list of medical AI benchmarks, including @SophontAI's very own Medmarks suite :)
Priyanka Shrestha@pri_the_ninja
Current map of medical AI benchmarks: 🧠 CLINICAL JUDGMENT • SCT-Bench tests whether models revise clinical judgments appropriately as uncertain evidence changes. • CPC-Bench tests complex diagnosis, next-test selection and medical-evidence retrieval. • Sequential Diagnosis tests whether models ask the right questions and order the right tests before reaching a diagnosis. These capabilities are increasingly consolidated through MAST, which currently includes CPC-Bench and several benchmarks below. @drethangoh @vishnuravi @pranavrajpurkar @StanfordMed @harvardmed 🛡️ SAFETY + COMMUNICATION • First Do NOHARM v2 tests whether models avoid harmful recommendations and critical omissions in clinical management. • The HealthBench family, including HealthBench Professional which is where the graph is from, tests realistic patient and clinician conversations using case-specific physician rubrics. @OpenAI @thekaransinghal • MedRiskEval, including PatientSafetyBench, tests patient-facing risks such as unsafe advice, misinformation, overconfidence and bias. @MSFTResearch 👁️ MULTIMODAL REASONING • ReXrank Mini tests interpretation and report generation from radiology images. @pranavrajpurkar • IMCBench tests safe, accurate reasoning in image-grounded, multi-turn medical conversations. @AmazonScience • AgentClinic tests clinical agents that must interact with patients, collect multimodal information and use tools. @npjDigitalMed 🧰 EHR + AGENTIC CARE • MedAgentBench v2 tests whether agents can retrieve information and complete actions inside a FHIR-based EHR. @AndrewYNg @james_y_zou @NEJM_AI • PhysicianBench tests long-horizon physician tasks grounded in longitudinal patient records. • LongMedBench tests memory and temporal reasoning across repeated visits and evolving treatment. • EHR-Complex tests executable reasoning over real clinical databases. 🏥 BROAD CLINICAL WORKFLOWS • MedHELM tests clinical decisions, documentation, patient communication, research and administration. @StanfordHAI @StanfordMed @NatureMedicine • MedMarks provides an open benchmark suite spanning medical reasoning, calculations, information extraction and EHR tasks. I’m sure I missed some, lmk what to add!
English
@RomiLifshitz you should write a thread explaining this further
English
Guys AI proved the Closest Vector Problem is computationally hard to approximate, this reinforces the security of lattice-based post-quantum cryptography. Huge ‼️
Sebastien Bubeck@SebastienBubeck
yes, nonsofic groups exist: this statement is one of many new beautiful results proved by Astra, our next major model. We're releasing 10 such Astra proofs, complete with lean certificates and CoT walkthroughs for each of them. The results are wide-ranging, from von Neumann algebras (disproof of Connes' Rigidity Conjecture) to better bounds for high dimensional sphere packing, for circuit complexity, for monochromatic triangles in multicolored graphs, and more. More thoughts here: openai.com/index/ten-adva…
English
kinda wild man...
Sebastien Bubeck@SebastienBubeck
yes, nonsofic groups exist: this statement is one of many new beautiful results proved by Astra, our next major model. We're releasing 10 such Astra proofs, complete with lean certificates and CoT walkthroughs for each of them. The results are wide-ranging, from von Neumann algebras (disproof of Connes' Rigidity Conjecture) to better bounds for high dimensional sphere packing, for circuit complexity, for monochromatic triangles in multicolored graphs, and more. More thoughts here: openai.com/index/ten-adva…
English
thank you Jensen Huang and people nitpicking over Twitter vs. X for the substantial payout this period :)
Tanishq Mathew Abraham, Ph.D.@iScienceLuvr
Oh my God Jensen Huang is now on Twitter!!!! And what an incredible first post too!
English
Tanishq Mathew Abraham, Ph.D. retweetledi
Ask and you shall receive. IndieMath: a crowdfunding microsite for Fable to solve open math/CS conjectures.
I've contributed the first $10,000 for Fable to work on disproving a handful of CS conjectures: indiemath.ai
Patrick McKenzie@patio11
I’m surprised nobody has put up a microsite backed by a for loop and token spend for all outstanding “named” problems in mathematics.
English
Can someone explain to me what this means?!
netsky remix🌎@iron_redux
🚨🚨Breaking news from the math world! Open AI just proved nonsofic groups existed! GPT 5.6 Sol continues to replace mathematicians!!!
English
As Leopold's macroecon classmate during freshman year (kind of funny considering recent events), I can say that you shouldn't count him out
TBPN@tbpn
Here is the full letter Leopold sent to his LPs last night. Rumors of his demise are greatly exaggerated.
English
Tanishq Mathew Abraham, Ph.D. retweetledi
