Mark Weber retweeted
Our DFlash 2 just got its driver's license 🚗
@NVIDIADRIVE's new Alpamayo 2 Edge uses it to think faster, right inside the car. Huge congrats!
Want to drive even faster? Check out FlashDrive, our #CoRL2026 paper 👇
Meet NVIDIA Alpamayo 2 Edge, a 2B-parameter reasoning VLA model designed for real-time, in-vehicle deployment on NVIDIA DRIVE AGX Thor.
Generate driving trajectories and causal explanations powered by adaptive reasoning.
Learn more 👉 nvda.ws/4k3vG7m
Try it out on @huggingface 👉 huggingface.co/nvidia/Alpama…
Mark Weber retweeted
The AI industry has a massive compute efficiency problem. At Physion Labs, we think it's time to challenge it.
Today, we're sharing the technology behind NexTron, our world-intelligence training engine.
Our benchmarks on matched hardware:
⚡ Up to 20.9× SFT throughput vs. DeepSpeed (H100)
⚡ Up to 6.6× SFT throughput vs. Megatron (H200)
⚡ Up to 7.2× SFT throughput vs. Megatron (B200)
⚡ Up to 3.1× end-to-end RL throughput vs. Miles–Megatron
🔥 180B model fine-tuning on just 8 H100 GPUs.
Same hardware. Dramatically higher throughput. Comparable model quality.
We believe the next frontier isn't just building bigger GPU clusters.
It's unlocking more intelligence from every GPU, every watt, and every dollar spent on training.
#PhysionLabs #NexTron #WorldIntelligence #AIInfrastructure
Mark Weber retweeted
Claude has discovered a previously unknown enzyme system hidden in the DNA of bacteriophages. Beside the enzyme’s gene sits a long array of repeating DNA—a structure that looks somewhat similar to CRISPR.
We don’t yet understand what this system does, but only a handful of known systems share its features, and all of them are able to cut, copy, and paste DNA. Historically, the discovery of such programmable systems has helped revolutionize medicine. CRISPR, for instance, is now the foundation of genetic medicines. But it will take much more work to learn what this system does, and whether it can be put to similar use.
Read more: anthropic.com/news/claude-di…
Qwen3.8-27B at 144 tok/s on an M5 Max MacBook Pro ⚡
Meet Inco Splash: our open-source inference engine, built around the model and around Apple silicon.
Up to 3× the decode speed of Ollama, 2× oMLX, and almost 4× when an agent fans out into sub-agents.
Mark Weber retweeted
i work at nebius, so take the following with that in mind.
today @PalantirTech named @nebiusai its preferred sovereign ai infrastructure partner.
most people will screenshot the logos (and this is a big one). The real story is where the compute sits.
most “sovereign AI” talk still means a region dropdown: same closed model, different flag on the rack.
but that is not this.
sovereign ai, in this deal, is two halves that finally touch:
1. who is allowed to touch the data and the model
2. where the model actually runs
palantir’s half is the authorisation and isolation layer, aip, ontology, foundry, apollo. Who can see what, which model can touch which data, how an org keeps the advantage it trains instead of leaking it into someone else’s frontier system.
what a lot of their commercial customers still lacked was the other half: ai-native compute and inference that lives inside that perimeter, not bolted on as a separate hyperscaler account.
that’s the half we build.
After the integration period, eligible Palantir customers get our compute and inference endpoints inside the palantir enterprise perimeter. open models, adapted on their own data, for their own domain. Control of compute, data, and the model that comes out the other side.
Karp put it cleanly: our infrastructure powers your ability to run your own models under conditions you control. ontology + infra undergird the sovereignty partners are demanding.
Arkady’s complementary line is the one i hear from customers every week: they need performance at scale and control. most buyers have been forced to pick one.
what this does for palantir:
the sovereign ai operating system stops ending at software. Preferred infra inside the perimeter is how you sell sovereignty to commercial orgs that never had a clean option.
-> their distribution. our runtime. one boundary
what this does for us:
it puts capacity where the hard buying decision already happened. Not only startups chasing gpus, enterprises that already standardized on foundry/aip and were still stuck on where the tokens actually run.
Preferred partner status plugs ai-native infra into a sales motion that already exists. Training and inference land next to the data loop that lets open models beat generic closed ones on a specific domain.
We’ll work together to bring new capacity online faster, including modular data center deployments at sites where power is already available.
not infinite cloud, power-first capacity. The physical constraint named inside a software partnership.
Sovereignty without isolation is cosplay
Isolation without compute is a brochure
Compute without power is a waitlist
today is the claim that those three can be sold as one stack
We are partnering with Palantir to bring trusted AI cloud infrastructure to Palantir’s commercial customers.
Palantir has named Nebius its preferred sovereign AI infrastructure partner.
Nebius compute and inference endpoints will run inside the Palantir enterprise perimeter – giving customers control over their compute, data, and models.
The partnership is based on a shared vision that open models, continually adapted using customers’ own data, can deliver better domain-specific intelligence while improving control and security.
Read the full press release: nebius.com/newsroom/palantir…
Mark Weber retweeted
GPU residual value update:
- A100 $4,956 (-12.0% YTD)
- H100 $20,308 (+3.6% YTD)
- B200 $71,057 (14.4% YTD)
A month on, the picture has not much changed. A100 ( launched in May 2020) has been roughly flat near $5k since late 2025. H100 and B200 have in fact both appreciated as the rising GPU rental income more than offset time decay! That is not a 2–3 year scrap curve!
GPU financiability has increasingly become the central question for the AI buildout: credit, not chips or power, is what stalls smaller builds. Banks still often mark GPU residual to zero after three years of straight-line depreciation.
Our estimates are going-concern value or what the GPU should be worth if it keeps running. "Zero after three years" is the wrong prior for that number. A six-year-old A100 still printing ~$5k is the living proof!
As @JensenHuang argued in his essay, we believe that GPUs should be increasingly thought of as financeable capital assets with stable cash flows coming from AI inference.
Based on our residual fair value estimates, A100 stopped depreciating since late 2025 as rising rental income offset time decay of value. Meanwhile, the H100 and B200 chips have both meaningfully appreciated in value in 2026 as a result of the strong increase in GPU rental rates!
We are still learning when it comes to the question of economic lifespan of GPUs. It certainly doesn't appear to be 2-3 years as some seem to casually assume. The NVidia A100 chip was released on May 14, 2020, well over 6 years ago and its rental rates are still holding steady after a significant run-up in 2026!
Mark Weber retweeted
GLM-5.3-Flash, meet DFlash 2. Flash² — 2.8× faster!
huggingface.co/incoai/GLM-5.…
More to come (really) soon!
GLM-5.3-Flash is fast. DFlash 2 makes it faster!
Weights are up: huggingface.co/incoai/GLM-5.…
Mark Weber retweeted
Y Combinator CEO, Garry Tan, took the stage for 42 minutes at Startup School 2026 and explained how to build your own personal AGI better than any paid AI course.
This is what he told the room:
1. The leverage is in your context, not the model.
Tan watches hundreds of founders use identical models every batch.
"There are 2x people and there are 100x people who are using the same Claude. Same weights, same context window size, same API. But the leverage is not in the weights."
The gap between users is now bigger than the gap between models.
2. One person's output went up 400x.
In 2013 Tan shipped maybe 14 useful lines of code a day as a YC partner, dead on the median for programmer productivity.
"I did the math on my output, and I'm at about 400x what I did in 2013."
3. Agents run on a different working memory.
Humans hold 7 things in their head at once. Every org chart and checklist ever built is a patch for that limit.
"An AI agent holds a million tokens. That's about a thousand pages. Three Harry Potter books sitting open on its head all at once."
You're still running your week on tools built for the 7-digit brain.
4. Markdown is code now.
Tan's stack is mostly skill files: pages of plain English an agent can execute.
"If you can write clear instructions in English, you're a programmer. The compiler is a language model."
At YC, finance and events staff who never opened a terminal are building automations.
5. Your history is your moat.
Tan's agent runs on a personal wiki: about 220,000 markdown pages covering 25 years of email, meetings, notes and decisions.
"When my agent does anything, it does knowing everything I know. And that's the difference between an assistant and a colleague."
No frontier model has your context. That's the one asset nobody can replicate.
6. Never do one-off work.
Most people run a task with an agent, close the window and throw the learning away. Tan ends every task by having the agent turn what it did into a reusable skill file.
"If you have to ask for something twice, you failed."
Captured skills compound daily. Amnesia resets you to zero every morning.
7. Own your skill files before your employer does.
A skill file is your judgment, extracted and executable. The only question is who controls it.
"Own your skills because if you don't, your job becomes a skill file."
Files in your repo compound your career. Files in the company's repo run your judgment without you.
Watch it, then read the step-by-step guide on becoming an AI engineer.
Mark Weber retweeted
Can deep learning work on small data with far more features than samples? We present PLATO: a method that achieves the state-of-the-art on such datasets by using prior domain information! neurips.cc/virtual/2022/post… 🧵
Published in #NeurIPS2023 with @ren_hongyu @kexinhuang5 @jure
Mark Weber retweeted
📢 Introducing Voyage AI @Voyage_AI_!
Founded by a talented team of leading AI researchers and me 🚀🚀.
We build state-of-the-art embedding models (e.g., better than OpenAI 😜).
We also offer custom models that deliver 🎯+10-20% accuracy gain in your LLM products. 🧵
Mark Weber retweeted
Tech from my team @MITIBMLab was just part of a demo in our CEO's opening keynote at #IBMThink 🎉—the result of a fantastic collaboration led by @rpanda89 and @yoonrkim
Mark Weber retweeted
The inimitable @kayellbee demoing our new watsonx.ai foundation model tooling for attendees at #IBMThink
Mark Weber retweeted
Do you want to estimate the density ratio accurately but are dissatisfied with the results?
Mark Weber retweeted
Happy to announce our #iclr2023 paper Contrastive Audio-Visual Masked Autoencoder CAV-MAE
Paper: openreview.net/forum?id=QPtM…
Code (incl. checkpoints ): github.com/YuanGongND/cav-ma…
Details below 👇
#computervision #machinelearning #ai
@MIT_CSAIL @MITIBMLab
Mark Weber retweeted
How did we develop the ResNet-50 training procedure that blew away the competition at MLPerf? Check out @leavittron's deep dive into the methodology behind it. The most important part: this methodology is repeatable. Stay tuned for BERT, DeepLab, GPT, and more to come!
Check out our deep-dive blog post on the Mosaic ResNet training recipe. See the details of our observations, and how to reproduce these results for your needs. We're able to achieve 7x faster training for ResNet-50, and so can you! #EfficientML
mosaicml.com/blog/mosaic-res…
Mark Weber retweeted
What’s next in AI? Flexible, reusable models known as foundation models that can be applied to nearly any domain or industry task and help limit the energy and data needed for training. Learn more: ibm.co/3yld62N
Mark Weber retweeted
Today, IBM announced it is working with @HSBC to accelerate #QuantumComputing readiness. Together, they will explore the use of quantum computing to reshape the way the bank tackles challenges like sustainability, pricing optimization, risk, fraud & more. newsroom.ibm.com/2022-03-29-…
Mark Weber retweeted
Joyful researcher moment: my first paper in adversarial #ML just passed 1K Google citations🙂I view the number as a metric of inspiration to others instead of achievements. It starts with a simple idea with @huan_zhang12 to connect zeroth-order opt with effective black-box attack




































