I've used LLMs to get the distributed Sprite OS running again, though I've yet to release it, and I'm not sure what that release will look like, if it ever happens.
One thing I really don't get, and I really hope I never do, is leave the front page/README/etc so ... Claude-ified (or GPT, plausibly, I guess):
native SCSI and ethernet drivers, proven on real hardware
Quake — runs; a benchmark more than a game, for now
running under emulation, real hardware next
The grimoire records both, confidence-tagged
I am becoming probably too okay with LLMs, and I'm probably about as "forgiving" of this stuff as anyone, but man, it's disheartening to see that for something as much of a passion project as "Amiga Unix", seemingly no-one bothered to write a page in a human voice for it. Maybe it's LLM-written, maybe a hybrid thing, I dunno. But at best it sounds like "Apple product release", IMO.
I know others will say the same about seeing LLM usage at all, and yeah... I don't really have a defense. It just feels like the "bare minimum of decency" to, well, at least make the front page less LLM-sounding.
Can't argue with you there. There's a ton of cool projects that people have vibecoded, and exist in domains where the fact that it's possibly insecure/poorly optimised/unmaintainable by humans doesn't really matter much. e.g. porting retro systems to, or writing new utilities for MiSTer FPGA. I love the platform, and the more stuff I can run on it, the better. Bugs are annoying, but not critical. The community gets software that may have never existed otherwise.
But still, somehow, when I see a Claudish README, it instantly gives me the ick. Write about your project. Tell us why you built it. Talk about the problems the bot helped you work through. I'm not going to judge you for not being able to code it all up yourself - but at least write the README in your own voice and tell me why you cared enough to do it.
Generally being succinct goes hand and hand with being zero knowledge.
The proof must be MUCH smaller than the whole computation or even the inputs (otherwise, just provide the uncompressed image!).
And the proof has to resist forgery.
So going for succinctness and at least computational soundness gets you to zero knowledge 'for free'.
Plus as the dead sibling comment notes: You may want disclosed modifications like a crop or redaction where the committed information remains private but you want to keep the proof.
Large lab "hacking" is only for the purpose of pushing competition suppressing doomer stories. You can tell by the fact their security is fine where it counts: keeping their weights and internal execution harnesses trade secret.
Did you see the account of some group getting a bounty payout of $6500 after using an exploit to get access to an employee’s github account and create a issue or PR (Idr which) on a private repository?
Seems like they could have potentially gotten access to the weights if they weren’t concerned about not doing crimes.
A better benchmark might be asking the LLMs to write GO ai and then comparing that-- the issue is that there will be a HUGE difference in performance that depends purely on this game being in the LLM's training... but training a general LLM to directly play these games would be a waste of capacity and shouldn't be encouraged for benchmaxxing sake.
Programming an engine OTOH is a skill that is more general and they should all have.
Might be useful to have the target of the engine be some specific virtual machine that gets a strict cycle budget-- e.g. execution runs so many cycles, and result is read out of a specific memory address at the end (or when it terminates early).
This is a great idea. It could learn from its mistakes, repeat things that work, abstract complex situations. I would even want humans monitoring the project.
Playing these games autoregressively isn't even the right way to use the LLM for this task (unless it was trained to do so...). It's somewhat like having a creative writing bechmark but requiring that all the input/output be base64 encoded. It can do it-- but no guarantees on the results!
And it's also just bencmaxxing bait: you can get a huge improvement on the task by RLing on it, but make no improvement on anything else. Doing so would just waste model capacity.
If you could tell that every LLM was equally not being exposed to the task then you could justify it as a test of abstract reasoning, but you can't. So it ends up on how much go transcripts ended up in the training, which is ... not a very interesting metric.
It'll be interesting when people start applying similar effort that is going into LLM search for security vulnerabilities to finding conventional reliability/performance bugs.
Say you have some code that should not be reading the initial state and is buggy if it does. Without zero-init, valgrind and msan will give you an immediate and false positive message that your code is wrong-- or forget dynamic analysis: the compiler can often statically tell you that the code will use an uninitialized variable. Zero initialize it and you lose that signal.
They never shipped the zen 5 epyc big cache chip (e.g. 9685X). :( I'd hoped they pulled it to produce a HBM integrated chip instead, but they didn't do that either.
I'm confused by a number of comments here and the article seemingly taking a position that a $250k lab instrument is required.
For one a similar instrument can be constructed from surplus parts for far less. Secondly, it's a single bit flip required. Now knowing the the technique works, a harness could be built that attempts it scattershot without the precise targeting and just has to try a lot of times. Using a different stimulus, e.g. xray it might well be possible without deencapsulating the part.
And be super-bankrupted by patent litigation from Apple. I don't think they're worried.
After all, they successfully threatened Adobe with spurious patent litigation unless they joined w/ apple in illegally fixing wages.
You don't think a criminal like apple would absolutely decimate any competition given the opportunity? They didn't hold back when it was a unambiguous crime, they surely wouldn't if it was merely bad for the world.
As in ... "the AI brought it back to life, but it came back /wrong/".
reply