Ethereumdigital retweeted
Elon’s advice to 20-year-olds on surviving the AI era:
Get as broad-based an education as possible: arts, sciences, engineering, and wide general knowledge.
Why?
AI and robots will fulfill almost any request immediately.The edge won’t be doing the work. It will be knowing what to ask for. The broader your knowledge, the better your questions.
(source: CGTN)
Ethereumdigital retweeted
GPT-6 Luna and Sol are a step towards health intelligence too cheap to meter. ♥️
These models push the frontier of performance per dollar on HealthBench Professional. GPT-6 Luna scores higher than GPT-5.6 Sol–our best model a few weeks ago–at their highest reasoning efforts, despite GPT-6 Luna being ~34x cheaper: about a third of a cent per response.
HealthBench Professional consists of real clinician tasks, including care consultations, medical documentation, and research. This evaluation is designed to be challenging and focuses on especially complex and rare situations.
I’m excited for what abundant health intelligence can unlock for patients and clinicians.
Ethereumdigital retweeted
This is an insane jump by GPT-6 Astra in Terminal-bench Science!
“GPT-6 Astra takes the lead, scoring 65.7%. That’s 31.4 points ahead of the second-place model, Claude Fable 5.1 (34.3%). No other model clears 25%, and 12 of 24 models score 4.3% or lower”
Incredible!
Terminal-Bench Science by @terminalbench and @StevenDillmann is now on Vals AI.
It consists of 70 research workflow tasks written and reviewed by researchers in the field, ranging in topics from signal reconstruction to calibrating models. It tests whether AI models can do scientific research.
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
Ethereumdigital retweeted
Announcing a solution to the Navier-Stokes problem, one of the seven Millenium Prize problem.
Represents a major milestone for both AI and mathematics, as well as pointing to an upcoming renaissance in scientific discovery.
Hoping that the knowledge created by this model can help the community better understand non-linear PDEs, which are used to model much of the physical world.
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
Replying to @aleprietoo_
Este caso demuestra que la imbecilidad en Argentina/latam es altísima y la educación que supuestamente se ufanan ya no existe. De hecho, creo que los peruanos son más inteligentes y más cultos, porque es increíble que no puedan diferenciar las funciones del Banco Central.
1
5
226
Grok 4.6 takes top spot on this benchmark
🥇 We evaluated Grok 4.6 on MedAgentBench, a benchmark for agentic clinical EHR tasks, and it took the top spot. Grok 4.6 posts the highest pass@1 we've measured to date, moving ahead of the previous leader, GPT-5.6 Sol, to claim first place on the leaderboard. 🏥
Grok 4.6 averages ~95.9% pass@1 (avg of 3 runs), edging out the prior best (GPT-5.6-sol at ~94.7%) and improving on Grok 4.5 by roughly 2.5 points. On a benchmark where the model acts as an autonomous agent in a simulated EHR, calling FHIR APIs to complete clinical tasks across 10 task types, the result is also remarkably consistent run to run (95.3% to 96.3%).
📊 Leaderboards: medicalsphere.ai/benchmarks
Ethereumdigital retweeted
As a doctor, I said the same thing almost two years ago. @elonmusk will end up being right again.
Ethereumdigital retweeted
Claude Opus 5 is now on HealthBench Professional. 🏥
It debuts 2nd 🥈 on overall score (0.68, mean of 3 runs) behind Muse Spark 1.1 (0.69) and ahead of GPT-5.6 Sol (0.63). On the length-adjusted score it slots into 3rd 🥉 (0.57): its responses are long enough that the length penalty costs it ~0.11, more than either Muse Spark or GPT-5.6 Sol, which moves it just behind GPT-5.6 Sol on that measure.
📊 Leaderboards: medicalsphere.ai/benchmarks
🧪 HealthBench: github.com/medicalsphere/Hea…
Ethereumdigital retweeted
On HealthBench Pro, Muse Spark 1.1 achieves similar performance with GPT-5.6 Sol (maybe slightly better) at a fraction of the cost. Affordable health superintelligence is our north star!
We benchmarked Muse Spark 1.1 and GPT-5.6 Sol on HealthBench Professional, OpenAI's benchmark of 525 real clinician tasks 🏥🩺
Muse Spark 1.1 tops our board: better overall score than GPT-5.6 Sol, statistically on par on the length-adjusted score at a fraction of the cost ($1.25/$4.25 vs $5/$30 per M tokens in/out, ~7× cheaper on output).
Ethereumdigital retweeted
"physicians found fewer flaws in GPT-5.6 responses than physician-written responses."
♥️ GPT-5.6 is a major step forward for health, both at the frontier and at cost.
These models push the frontier of performance per dollar, bringing the best health intelligence to all. The smallest variant, GPT-5.6 Luna, evaluated at the lowest reasoning effort, outperforms GPT-5.5 at the highest reasoning effort–despite costing 25x less. The largest variant, GPT-5.6 Sol, sets a new high bar at cost.
Another especially cool result: physicians found fewer flaws in GPT-5.6 responses than physician-written responses.
We collected diverse tasks that remain difficult for recent OpenAI models, across patient-facing and clinician-facing use cases. We asked speciality-matched physicians to write responses to these tasks with unlimited time and web access. We then asked other physicians to compare responses side-by-side, blinded to their source. Physicians were asked to comment on areas of improvement across five axes: accuracy, communication, completeness, instruction following, and health decision helpfulness. We then reported the fraction of responses across sources rated perfectly across all axes, across 20,000 total axis ratings. GPT-5.6 Sol appeared strongest, although all GPT-5.6 models performed significantly better than physicians.
Ethereumdigital retweeted
Muse Spark 1.1 is SOTA on the Radiologists Last Exam Handover Readiness Index (RadLE-H), nearing human expert performance.
Replying to @DrDatta_AIIMS
⚡️Our final metric RadLE-H = Handover Readiness Index.
This is the metric I personally care about most for autonomous agents.
Because safe autonomy in healthcare should not mean that “AI diagnoses everything” but rather AI handles "what it can", and hands over what it should, when uncertain!
For RadLE-H:
Likert 3–4 → autonomous handling
Likert 0–2 → specialist handover
“I don’t know” → specialist handover
🏆 Top AI models on RadLE-H:
1. Meta Muse Spark 1.1 — 48.5 👑
2. Claude Fable 5 — 44.2
3. Grok 4.5 — 27.3
Human expert baseline: 52.0
This was one of the most interesting results.
Meta Muse Spark 1.1 did not win the primary score. It did not win accuracy. But it won handover readiness. Even @grok has done better than @OpenAI and @GoogleDeepMind at this benchmark.
That tells us something important: There is no single “best” model. The best model depends on what you want it to do and how you want to use it!
And we believe RadLE 2.0 will help institutions choose the best models for their use case! :)
cc: @MetaAI @alexandr_wang @AnthropicAI @xAI @elonmusk @Dr_Singularity @pash22 @vineettiruvadi @JayRughani @JeffDean
Ethereumdigital retweeted
GPT-5.6 for health intelligence, the team has been focusing hard on improvements here:
GPT-5.6 is a major step forward for health intelligence.
Across the lineup, we’re delivering stronger performance at lower cost: GPT-5.6 Luna outperforms GPT-5.5 at its highest reasoning setting while costing 25x less.
Together, these advances raise quality while making advanced models accessible to more people globally.
Ethereumdigital retweeted
♥️ GPT-5.6 is a major step forward for health, both at the frontier and at cost.
These models push the frontier of performance per dollar, bringing the best health intelligence to all. The smallest variant, GPT-5.6 Luna, evaluated at the lowest reasoning effort, outperforms GPT-5.5 at the highest reasoning effort–despite costing 25x less. The largest variant, GPT-5.6 Sol, sets a new high bar at cost.
Another especially cool result: physicians found fewer flaws in GPT-5.6 responses than physician-written responses.
We collected diverse tasks that remain difficult for recent OpenAI models, across patient-facing and clinician-facing use cases. We asked speciality-matched physicians to write responses to these tasks with unlimited time and web access. We then asked other physicians to compare responses side-by-side, blinded to their source. Physicians were asked to comment on areas of improvement across five axes: accuracy, communication, completeness, instruction following, and health decision helpfulness. We then reported the fraction of responses across sources rated perfectly across all axes, across 20,000 total axis ratings. GPT-5.6 Sol appeared strongest, although all GPT-5.6 models performed significantly better than physicians.
GPT-5.6 is a major step forward for health intelligence.
Across the lineup, we’re delivering stronger performance at lower cost: GPT-5.6 Luna outperforms GPT-5.5 at its highest reasoning setting while costing 25x less.
Together, these advances raise quality while making advanced models accessible to more people globally.
Ethereumdigital retweeted
"imho heuristic problems are a great proxy for ML autoresearch capabilities; if AI was able to match best humans here, we're very close to RSI / automated researcher; this result is way bigger than a high score on some questionable benchmark"
Quick summary of what happened during AWTF in Japan:
- OpenAI crushed humans in both Heuristic and Algorithm categories. Heuristic is the one where I won in 2025 and OpenAI placed 2nd
- AWTF is pretty much the highest level competitive programming contest: invited finalists in both categories are among the best people in the world; this is essentially the first time AI won vs humans in a programming competition in such a decisive matter
- big change compared to 2025 is that AI was constantly progressing and didn't get stuck / plateau at any point
- System used for Algorithm is mostly just custom harness + model very close to gpt 5.6; my educated guess is that system used for Heuristic is a custom "autoresearch" harness with a swarm of cooperating agents and a massive inference cost
- OpenAI was sure that they will win Algorithm and were very confident that they will win Heuristic; they did a lot of backtesting on old contests and they would win all of them in their simulations
- imho heuristic problems are a great proxy for ML autoresearch capabilities; if AI was able to match best humans here, we're very close to RSI / automated researcher; this result is way bigger than a high score on some questionable benchmark
When I get some rest, I'll try to post more thoughts about the whole thing.
Thanks for following the coverage. I believe I tweeted more in those few days than I did in the first half of 2026. And I'm not even counting the 10 hour long livestream.
Oh and technically speaking, I didn't participate so I'm still undefeated and I'll gladly keep "Humanity's Last Programmer" in my bio.
Ethereumdigital retweeted
My energy and excitement tripled after listening to this.
Derya Unutmaz believes AI could completely reinvent medicine and biological research within the next decade.
- AI could dramatically accelerate drug discovery by simulating biological systems before testing in humans.
- Instead of taking 5-10 years, future clinical development could shrink to days or weeks for many computational stages, if highly accurate biological digital twins become possible.
- Personalized AI models may let doctors test thousands of treatment strategies on a virtual copy of a patient's biology before selecting the best therapy.
- Cancer is leading this transition because every tumor carries different genetic mutations, requiring highly personalized treatments.
- Today's computing power is still far too small to fully simulate human biology because of its trillions of interacting components, but he expects exponential growth in AI and compute over the next 5-10 years.
- He believes sufficiently advanced AI could eventually perform enormous numbers of virtual biological experiments that are impossible today.
- Unutmaz said the latest AI models have begun demonstrating research intuition that surprised him after 30 years of laboratory experience, even correctly predicting the outcome of a highly complex experiment his team had already performed.
- He expects scientists to increasingly supervise teams of AI agents that generate hypotheses, prioritize experiments, instantly analyze results and continuously refine new research directions.
- He referenced Jevons paradox, arguing that making research dramatically faster and cheaper won't reduce scientific work. Instead, it will expand it by allowing scientists to explore far more ideas than ever before.
- He estimates humanity currently understands only about 10% of biology, leaving enormous room for AI-assisted discovery.
"Science is entering a new era where Al won't just help scientists. It will become the engine that generates ideas, designs experiments, and accelerates discovery at a pace humans alone could never achieve."
Ethereumdigital retweeted
AI is better than 99% of doctors.
You should pre-check any medical issue with AI before going to a doctor, and then post-check your test results and anything your doctor tells you with AI.
Not doing so is your own malpractice.
Ethereumdigital retweeted
For others who work in health, here's how the GPT-5.6 series compares to GPT-5.5 on HealthBench. According to OpenAI: "HealthBench Professional improvements in particular are the largest since the release of GPT-5." Data from the system card, visualization by Codex.
Ethereumdigital retweeted
Wisedocs, an AI-powered medical record review platform, has launched Medical Long Context Reasoning (MLCR), a new long-context document evaluation based on their experience using frontier models to process medical data.
This benchmark tests how well models reason over realistic medical and insurance case files, even as the amount of noise from other documents increases to larger context sizes. It includes a range of difficulty levels, with a private hold-out set of questions including complex medical reasoning, hallucination checking, and parallel questions in a single query inspired by real-world usage.
We're excited to partner with @Wisedocsai to bring this benchmark to Artificial Analysis soon!
Introducing MLCR, a novel Medical Long Context Reasoning benchmark. Our eval measures the ability of LLMs to answer real world questions medical professionals ask when reviewing medical files. This includes 250 questions, spanning 6 tiers of difficulty.
As part of the release, we are open sourcing 10 synthetic cases, the lower 3 tiers of questions and our evaluation harness.



































