johnsutor

Maybe it's a play on the abysmal uptime of Claude Code?

Not exactly new but newer, Polo and Pan are great.

I'm working on so101-nexus, an open-source sim-to-real stack for the SO-100/SO-101 robot arms where you can record teleop demos, behavior-clone a policy, then fine-tune with RL. The goal is to be very compatible with Gymnasium, MuJoCo, and LeRobot.

https://github.com/johnsutor/so101-nexus

There was a Joe Pera episode where they made a musical about this https://www.youtube.com/watch?v=_LEJSMm1qpM.

https://platform.openai.com/docs/models/gpt-5.2 More information on the price, context window, etc.

Brings me back to playing Kerbal Space Program

It looks pretty akin to mathacademy.com, which is fantastic for brushing up on math concepts.

https://bloomberg.github.io/foml/#home This course is my personal favorite.

This license is called a Fair-Source License https://fair.io/

Interesting idea. Similarly, recent work appears to have used MCTS to explore sequential multi-agent systems (see https://arxiv.org/abs/2410.10762, https://arxiv.org/abs/2410.17238).

There are plenty of "newer" technologies that closely mimic more "boring" tech too. For example, Svelte feels a lot like writing vanilla HTML and JS, Supabase is all built entirely on PostGRES instead of a shiny, noSQL database.

Seems like this is already being answered:

https://arxiv.org/abs/2407.10930 https://arxiv.org/abs/2006.04439

With the advent of "Kia Boys" and now this, it's a miracle people still buy Kias.

Or, you know, just create your own API for your platform and charge people per request to that.

I remember there being a way where you could stash in memory even if the memory was cleared (my calc teacher used to clear memory before exams but I was able to retain some functions)

Lots of recent development in the speech-enabled LM space recently (see https://github.com/ictnlp/LLaMA-Omni, https://github.com/gpt-omni/mini-omni)

Or just use Stream East...

Would love to have something like this for Svelte.

This would be great to have for the Triton language as well.

From the KAN repo itself, it appears they already have GPU support https://github.com/KindXiaoming/pykan/blob/master/tutorials/...

I know there are murmurs that synthetic data (i.e. using rendering software with 3D models) was used to train some generative models, including OpenAI Sora; seems like it's the only plausible way right now to get the insane amounts of data needed to capture such statistical regularities.

BERT isn't dead for smaller tasks (think NER, Sentiment Analysis) where low latency is needed.

How does this differ from Ragas? https://docs.ragas.io/en/latest/index.html

Last weekend, had a listening event at the Angel Orensanz Foundation, where they took the original photos for the 36th Chambers album release. Was a cool experience.

Long industrial technology, Short pre-industrial technology

They're only trained up to a certain point in time, so adding RAG should hypothetically allow such LLMs to access the most up-to-date information.

I've got a dust allergy and swear by the Levoit Core air purifier. It's crazy how much dust accumulates in such a short period of time (I live in a building with central air that doesn't have filters on the vents). I also try to open the windows for at least 10 minutes daily, even in winter.

Main thing I would be concerned about is safety (I’m an avid Volvo fan) but it appears to be (relatively) safe as well: https://microlino-car.com/en/service/faq/how-safe-is-the-mic...