The state of "Scientific AI". I know LLMs are accelerating at doing computer work, and I've experienced the acceleration of using LLMs to do science, but it's more along the lines of debugging and stitching together pipelines of classical in silico tools. I see a ton of value here along that entire pathway for LLMs, it's basically digital process development: take each step, make it better, repeat, repeat, repeat. But the ceiling are the unit operations themselves, which sure LLMs can improve the code of those tools next. But if science was held back by simply people not doing the same things faster, maybe this will really push us forward, but I have a nagging feeling of "is this it?".
I think what I'm really looking for is a model like GPT Rosalind, which has been steeped in "science" post-training, but with more randomness. Like, I think I'm looking for a GPT Mullis--Rosalind Franklin was careful, deliberate, and serious; Kary Mullis ate a bunch of acid, drove on the PCH, and invented PCR. Like, we need to invent psychedelics for LLMs. Some way to let them relax their weights, explore new pathways, and come up with some absurd ideas by connecting random ass dots from the natural world. I want the model to say, "hmm, that's weird". This isn't just changing the temperature, we're missing something deeper.
I think frontier science has always come from serendipity: a bright thinker, listening to a presentation after having stared at some small experiment, but having trained for years on "biochemistry" so the foundation and loose guardrails are there.
I don't know, I'm feeling adrift. Does this resonate with anyone else?
Yes! I had a mandibular advancement device for many years, but after about 10, my lower jaw stopped resetting all the way in the morning. I forced myself to find the right CPAP (nasal pillows with fabric elastic chin strap to keep my mouth closed). It's been about 5 years and my jaw finally is >95% back to normal (YMMV).
But by far the most critical part of getting CPAP to work was using an open source program to review the SD card raw data every morning, learning the interplay between max/min pressure, temperature, humidity, and ramp. Gaining access to the admin controls on my AirSense and fiddling it myself over weeks to find the right settings--this was the only way I made it work. The idea of waiting 6 months to make a change to the pressure was bonkers to me. Especially, for some reason, my brain is shit the next morning if I over oxygenate with too high max pressure overnight, worse than not wearing it at all!
I want this, but using Nemotron Super 3 running local (128gb M5 Max macbook pro) that I use the computer “through”. Does Goose AI aspire/do this? I just started working on this yesterday.
I love this! I really wanted to go down this road when my kids were younger, but the paucity of floppys and the low storage space made me go down the Avery business card print outs with RFID stickers on the back and a raspberry pi with an RFID reader inside. Of course, the author is using the floppys as hooks instead of as storage media...what a great idea. The tactile response and the art you can stick to them makes them ideal for this purpose.
I use Opus 4.5 and GPT 5.2-Codex through VS Code all day long, and the closest I've come is Devstral-Small-2-24B-Instruct-2512 inferring on a DGX Spark hosting with vLLM as an "Open AI Compatible" API endpoint I use to power the Cline VS Code extension.
It works, but it's slow. Much more like set it up and come back in an hour and it's done. I am incredibly impressed by it. There are quantized GGUFs and MLXs of the 123B, which can fit on my M3 36GB Macbook that I haven't tried yet.
But overall, it feels about about 50% too slow, which blows my mind because we are probably 9 months away from a local model that is fast and good enough for my script kiddie work.
Does anyone know how this actually was done? Like, did they export every frame as a PNG and then run them each one by one through the model? Or did they somehow "load" the video into the model directly (which then internally somehow steps through each frame?)
Do yourself a favor and study for both your technician and general at the same time (I’m assuming you live in the US). HF is exponentially more fun than just VHF/UHF.
I don’t disagree with your point, but I just would like to point out that there are over 100 known post-transcriptional modified RNA bases [1]. In fact, tRNA are more modified bases than canonical if taken as a whole. AND! the ribosome can’t function without all of its modifications. If I were to put money toward “targeting an RNA to make a drug” rRNA is where I’d aim…
The author says 36GB unified ram in the article. I run the same memory M3 Pro and LM Studio daily with various models up to the 30b parameter one listed and it flies. Can’t differentiate between my OAi chats vs locals aside from modern context, though I have puppeteer MCP which works well for web search and site-reading.
I think what I'm really looking for is a model like GPT Rosalind, which has been steeped in "science" post-training, but with more randomness. Like, I think I'm looking for a GPT Mullis--Rosalind Franklin was careful, deliberate, and serious; Kary Mullis ate a bunch of acid, drove on the PCH, and invented PCR. Like, we need to invent psychedelics for LLMs. Some way to let them relax their weights, explore new pathways, and come up with some absurd ideas by connecting random ass dots from the natural world. I want the model to say, "hmm, that's weird". This isn't just changing the temperature, we're missing something deeper.
I think frontier science has always come from serendipity: a bright thinker, listening to a presentation after having stared at some small experiment, but having trained for years on "biochemistry" so the foundation and loose guardrails are there.
I don't know, I'm feeling adrift. Does this resonate with anyone else?