manmal

26 ms·

31.
by manmal 1mo ago
One of the Codex employees on Twitter promised (threatened?) that the next gen of Codex will be cloud focused, and less local. I wonder now if that means it will be a more powerful GPT Work.
32.
by manmal 1mo ago
I got a 404 when heading to the examples.
33.
by manmal 1mo ago
I haven’t used Haskell much. I thought to try it with agentic help, but then read this post that its compilation times might make it unviable. But what’s your experience?
34.
by manmal 1mo ago
They have exactly one guy as the expert in this video and he contributes all the facts. That’s bias.
35.
by manmal 1mo ago
That video is quite biased. Just tell your clanker to critique it. It will point out that those numbers are quite cherry picked, and leaving out important details. E.g. the „GHG would reduce by only 2.6%“ study assumes that all crop product
36.
by manmal 1mo ago
My problem with memory is that it goes stale, and updates to facts are often not changing all locations of that fact. Such a system should make it easier to maintain a single source of truth, and versioning, no?
37.
by manmal 1mo ago
I guess you also get very high bandwidth that way? I‘m not sure that would come for free though.
38.
by manmal 1mo ago
Isn’t it just for development purposes? I see it as a web sandbox for iOS development.
39.
by manmal 1mo ago
They aren’t reasoning though, are they?
40.
by manmal 1mo ago
How do they like it, compared with 3.8 and DS4?
41.
by manmal 1mo ago
I didn’t expect an answer I can actually agree with - keep up the good work ;)
42.
by manmal 1mo ago
Sorry for the snark, but are you trying to cure cancer? What could possibly need 24/7 research in our domain, that doesn’t need your input every 30 minutes?
43.
by manmal 1mo ago
A Mac Mini is easy to match by a PC in terms of inference, and the PC will win without getting noisy.
44.
by manmal 1mo ago
1-2 RTX5090 will be better value than Macs because they have the memory bandwidth for somewhat fast local inference.
45.
by manmal 1mo ago
The upgrade to 6E was a big jump in my network (Wifi7 Unifi Wall AP). On the same spot, the M3 MacBook (6E) has basically zero lag over VNC, while the M1 MacBook (6) has a noticeable lag. It’s a bit mind blowing how big the difference is.
46.
by manmal 2mo ago
I‘m outsourcing all tedious searches to ChatGPT now. Anytime I try searching anything non-obvious, it seems to be buried in the Nth page. Both Google and DDG have this problem. Chat spares me the brain power of sifting through the endless s
47.
by manmal 2mo ago
> There is no framework here. The entire frozen layer is seed.py Slop language in the second paragraph. I refuse to read on.
48.
by manmal 2mo ago
Seems to me a bit insensitive or logarithmic. Fable is way worse than some of the others in this list, but only 10-20% higher score.
49.
by manmal 2mo ago
There‘s a lot of ancient biology at play in raising kids. You won’t get exactly this kind of feeling elsewhere. Doesn’t mean it has to be your kids though. You just need to be an important part of their life.
50.
by manmal 2mo ago
Why wouldn’t you use Luna for that? It’s super cheap.
51.
by manmal 2mo ago
Yeah. Local agent sessions are not backed up in the cloud. And uptime is better with local models.
52.
by manmal 2mo ago
Isn’t the fact Fable is more expensive than Sol-Max by multiples already an indication that Sol is way smaller?
53.
by manmal 2mo ago
Are you using the SOTA models at very high reasoning during planning? IME that makes a LOT of a difference. I‘d also never let them just rip into the architecture, but always push back and ask for alternatives first. Once the overall plan i
54.
by manmal 2mo ago
I do spend 5x more tokens on planning and reviewing, than for implementation. But architecture is still nothing I can delegate.
55.
by manmal 2mo ago
> At a certain speed point, you're able to move from request -> response to a cascade of tool calling and "subagents" That also needs server class hardware though. A phone won’t happily service the insane amount of IO,
56.
by manmal 2mo ago
I think that only works if you have checkpoints where all that reasoning can be checked against reality. Otherwise you get an army of armchair experts. LLMs are hilariously bad at home improvement advice btw, where reasoning alone won’t get
57.
by manmal 2mo ago
About your first example, isn’t the butterfly effect preventing this from being useful? One agent in your simulation decides to sell, and starts an avalanche, that won’t happen in reality?
58.
by manmal 2mo ago
Problem is, there exists no judge model that will really pick the same winner that you would.
59.
by manmal 2mo ago
Errors compound, and making 1000 wrong decisions per hour, will not result in something useful. Maybe you‘ve tried setting up guardrails for good design or architecture at some point? I think it’s simply not possible to do that. It would ce
60.
by manmal 2mo ago
That’s their API with extra steps, or am I missing something? That was always faster.

More ›