skerit

4 ms·

31.
by skerit 3mo ago
But LLMs aren't just shuffling words around. Words (tokens) are the "human interface" at the edges. Text input gets embedded, then there's this huge latent-space computation in the middle, and only at the tail end does t
32.
by skerit 3mo ago
I immediately saw the Ghidra one and was thinking: huh?
33.
by skerit 3mo ago
> the improvements are getting smaller and smaller The AI haters have been saying this for 2 years now.
34.
by skerit 4mo ago
And here I was thinking we'd be getting Sonnet 5. Or maybe even poor old Fable back. Not this junk.
35.
by skerit 4mo ago
Which AI bubble? The stock bubble one or the imagined "one day all the LLMs are going to disappear from the face of the Earth" one?
36.
by skerit 4mo ago
Was this not the case earlier? I've had it stuck in loops from day 1, even when Gemini 3 came out. I've never been able to actually use it for anything because it goes absolutely mental after a handful of turns.
37.
by skerit 4mo ago
Yeah, that approach makes the most sense.
38.
by skerit 4mo ago
Oh yes, of course. I was talking about reverse engineering the code only. Requiring the official assets is a no-brainer.
39.
by skerit 4mo ago
> it would be extremely easy for the rightholders to claim that both agents have most certainly ingested the binary during their training phase Ingested the binary?
40.
by skerit 4mo ago
I'm currently reverse engineering a few games too. It's quite easy with AI now. But I'm worried about the legality of it all. Any thoughts on this?
41.
by skerit 4mo ago
I've been working on something like this too, for quite a while! Though I'm trying to get a non-quadratic-attention LLM (or SLM) up and running. And anyway, I think the most important thing is dataset quality. Dumping in whatever
42.
by skerit 4mo ago
I've been creating my own little from-scratch LLM for months now with Claude's help. I can safely say I learned a thing or two along the way.
43.
by skerit 4mo ago
> Burned $2K to see how it will perform on frontend tasks and backend tasks Burned $2K on some kind of enterprise account or ... ? Why not just get a $200 Max Pro account? While I'm loving the output of Fable 5, I will *never* pay t
44.
by skerit 4mo ago
I've seen Opus do some incredibly token-costly things before too. In fact after most sessions I ask it about which tools it used often, which tools could be simplified/made less verbose, could be "combined" into one, ...
45.
by skerit 4mo ago
Fable 5 on medium is amazing. It's handling everything I throw at it I had _one_ instance where for some obscure reason it decided to fall back to Opus 4.8 and Opus IMMEDIATELY fucked it up and implemented a super obvious feature in a
46.
by skerit 4mo ago
That's strange... I've been tinkering with a little LLM-from-scratch project for a while now, and Fable is just continuing it without a problem
47.
by skerit 4mo ago
It's a very nice bump, but it is in no way worth all the hype of the past month.
48.
by skerit 4mo ago
> although opus 4.8 card had mentioned an 'honesty upgrade' If I never see Claude say "I have to be honest" ever again I'll be happy.
49.
by skerit 4mo ago
Oh nice, it didn't flag the request? I feared any reverse engineering would become impossible because of the new safeguards.
50.
by skerit 4mo ago
Same here. Claude isn't perfect. It still makes a lot of mistakes. But whenever I try GPT-5.5 it's ten times worse, and Claude just has to clean up GPT's mess.
51.
by skerit 4mo ago
And then you have me, using gitmoji
52.
by skerit 4mo ago
> GPU compute for self-study Those suggestions they make for a B200 start at $4.99 an hour. Is that really required, for starting out? I've been tinkering with my own from-scratch LLM, but in the early phases I don't need anyth
53.
by skerit 4mo ago
Great and all, but what about the legality of it all? It wouldn't be the first time that a decompilation project gets shut down. Reverse engineering is (luckily) OK, but reproducing (or, well, releasing) the actual original code with t
54.
by skerit 4mo ago
I just use /loop now which gives it a motivational prompt to keep the fuck going. Goal can be used too, but for some things a simple loop is better
55.
by skerit 4mo ago
I noticed (and absolutely HATE) that Opus 4.7 likes to start any negative response with "I have to be honest" or whatever. It drives me mad.
56.
by skerit 4mo ago
It is. It's super easy to do so. You can send anything via the API.
57.
by skerit 5mo ago
You can use a modded client, everyone does, but there _are_ some protections. So you can get kicked, you just won't get banned.
58.
by skerit 5mo ago
2b2t is so interesting. It is a toxic place, yet you can find really nice things, builds, messages, ... It's fascinating.
59.
by skerit 5mo ago
Netsplits were fun. Especially if you were on the splitting part. You could get to know new people you got stuck with.
60.
by skerit 5mo ago
> We know this because your IP address was the first thing your device sent us. First paragraph, and I don't like this wording already. It's as if "my device" has any choice in the matter. And actually, it's the

More ›