pseudotensor

Will do!

Update: Microsoft just announced Microsoft-Decision-1 (a post-trained Qwen3.5-9B) and benchmarked it against H2O-Lightning-4B, among others: https://commandline.microsoft.com/microsoft-decision-1-model...

One note on their speed comparison: the JevBench board shows self-hosted models with an adjusted latency of "2x + 0.15 s (assumption, not measured)". Our measured p50 is 29 ms; the adjusted figure is 0.21 s. They report 85 ms for theirs.

Disclosure: I work at H2O.ai.

We released an Apache-2.0, open-weight 4B decision model that scores above Jev 1.13 on JevBench's composite score (72.5 vs 71.5) and is currently the top open model there: https://benchmarkheaven.com/jev-models . Newer models coming even larger than beat Jev in intelligence as well.

- Same contract as Jev: state + typed questions in, calibrated probabilities out, one forward pass, no generated tokens. - Your data never leaves your environment, and there's no per-call fee.

Weights, card and run instructions: https://huggingface.co/h2oai/h2o-lightning-4b

Easy game.

Protecting math jobs limits cancer progress even being a choice of what to pursue. You could choose to allow 24/7 surveillance, but why would you? At least with math progress society can choose what to allow from it, but slowing it down universally blocks all choices.

Tao et al. are effectively calling for diseases like childhood cancer to remain persistent for longer.

Physics and Biology will see major breakthroughs that WILL fundamentally alter our world. That is one key thing missing from alot of discussion here is the narrow focus on math (or parallels with software engineering). Doing well in math is key to doing well in physics and other sciences.

This is "just" an attribution and credit assignment problem. OpenAI could have done vastly better than they did at attribution. They should have spent another $10M just on attribution/credit research to annotate the contributions to the lean and paper and their blog.

"guise" doesn't mean they don't care. It means they are shadowing their concerns when in reality they have concerns primarily about what AI will do to their success in math and the credit they will receive in the rest of their lifetime -- i.e. their legacy.

Job protectionism for elite mathematicians under guise of caring about student development. The glory of the super smart math person will need to shift to more creative modes, just like art had to handle photography. Attribution is legitimate issue but should not stall progress as it is easy to address via the same research mechanisms that agents already do.

This is related: https://www.reddit.com/r/LocalLLaMA/comments/1fiw84a/open_st...

The idea is not silly in my view, I did something similar here: https://github.com/pseudotensor/open-strawberry

The idea is that data generation is required first, to make the reasoning traces. ToT etc. are not required.

New fully open source LLM chatbot and 7B, 12B, 20B models. Try it at https://gpt.h2o.ai/ .