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.
QR codes on cards would work as well, if I'm understanding what this project is. The floppy disk approach has some nostalgia maybe but seems quite fragile. I quickly learned to never let my kids handle CDs/DVDs (one of the worst physical media designs ever; they are totally unprotected) as they would quickly become damaged and unplayable. Floppy disks are at least sort of protected but the same idea applies.
I still have a large number of working CDs from when I, myself, was a kid. DVDs too but they were later and more durable.
I’ve always wondered what people are doing to them? Maybe I just got lucky. Maybe I was just careful with them. Maybe I don’t remember the ones that failed.
I don’t think kids are less careful now, although being screamed at for making the CD or record skip was probably a deterrent.
Some people really get the idea of only handling it around the edges. Lots of other people just handle them however they want and have no problems touching the media anywhere. Especially kids, which often don't have the cleanest hands at any given moment.
Lots of kids will handle them however they want. They'll pick them up with greasy, sticky hands right on the media section. They won't necessarily care about ensuring they're properly in the drive tray. They'll jam all kinds of things into the drive slots. They'll drop them on the floor and step on them, toss them in a toy box when told to clean their room, etc.
Obviously not all kids will be this way, but many will.
I don't think I can get my hands on a floppy drive, but I still have an ancient computer somewhere with a DVD player in it. While not as cool, I had been considering turning into a simple media station for the specific purpose of letting my kid pick what music to play or video to watch by herself, without needing a screen to navigate it.
Like you, it never occurred to me that I can also just use specific DVDs or CDs as hooks for videos to be streamed, or media downloaded on a hard drive. So that suddenly makes the whole project a lot more interesting, and possibly easier too.
Buying a large pack of burnable DVDs is a lot cheaper and sustainable than using SD-cards like other commenters suggested.
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.
The US ham test question pools are fully public. Your test will be a mixture of questions from the pool. HamStudy basically lets you churn the question pool, and then will offer explainer text / references to back up each question and correct answer.
I went on a vacation and used their phone app any time I was standing in a line. You can set it to just keep spinning through the questions, with a bias towards ones you're getting wrong.
You need to get 37+ correct to pass. Another way to think of that is you can get up to 13 wrong and still pass.
Within each category there are subcategories. "Antennas and Transmission Lines" for example has 8 subcategories. The 8 questions in "Antennas and Transmission Lines" are one from each of those subcategories. The question pools for these subcategories each have 10-14 questions.
If you compare to the closest corresponding categories/subcategories from the General and Technician exam you'll probably find that there are a few cases.
1. The Extra is just more of the same. It's not harder per se. "Commission Rules" for example.
2. The Extra goes goes deeper and also adds new material that is more advanced.
3. The Extra is in new territory.
If you get to the point where the Technician and General are going to no problem, then you will probably have no trouble getting to the point where case #1 is also no problem, and case #2 is also well in hand. It is #3 where you might have trouble.
But remember that you can get 13 wrong and still pass!
Pick say 10 subcategories that are in case #3 that look like they would be the hardest to get good at and just write them off.
For example in "Antennas and Transmission Lines" you might decide that the "Smith chart" subcategory, which has a pool of 14 questions, would take a lot of time to get good at. So skip it. That's 14 less potential questions you have to be prepared to answer, leaving more time to study for things in class #2 and the class #3 things that look most doable.
It doesn't cost extra to take the Extra test at the same session that you take the Technician and General tests, and there is no penalty for failing, so might as well go for it.
Huh, I never heard that one. Extra gets you more frequency privileges (nice not having to worry so much about band edges) but IMO the real benefit is being able to enjoy reciprocal operating under CEPT when traveling abroad.
Only the entry level license (Technician in the US, covering UHF/VHF) is substantially different from the German one, and it's also much more restricted. Germany in general is a much better country for radio, especially if you wanted to ever do high power broadcasting.
I think the one perk to the US is that the FCC has basically stopped caring about all but the most important frequencies. This makes HF particularly fun, since HF pirate radios are often the best listening stations in the entire RF spectrum. I have no idea what that's like in Germany, but I would imagine given the general ordnung culture and veneration of rules, German hams are much less tolerant of flagrantly unauthorized broadcast stations and your regulatory bodies are more proactive in shutting them down.
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?