Bolwin

4 ms·

31.
by Bolwin 2mo ago
Why is the title "Africa" and not Morocco
32.
by Bolwin 2mo ago
Screams it in fact
33.
by Bolwin 3mo ago
Jeez way to ruin of the few remaining joys of flying. Lock everyone in a tin can. You will experience reality through screens only and you will enjoy it
34.
by Bolwin 3mo ago
Llms use them a lot more than humans, including this blog post. Like all slop. There's a reason it's called slop and it's not because of restraint
35.
by Bolwin 3mo ago
I like this. It feels like more a person talking and less like a prepared speech.
36.
by Bolwin 3mo ago
For a fair comparison, you should compare to K3 (which AA has not tested yet unfortunately) and GPT 5.6 Sol also on medium or the closest equivalent
37.
by Bolwin 3mo ago
This is a lot of words to say "switch models instead of writing a plan file and starting a new session" which I do anyway. That said, it takes me a while to reads plans and often I'll take a break, by when the cache has expir
38.
by Bolwin 3mo ago
X.com is not publicly accessible. I wish people would stop using it as a source
39.
by Bolwin 3mo ago
Every time a competitor for acquired, sold out, or changed business models, Kilo made a snarky post. All companies are the same
40.
by Bolwin 3mo ago
Where are you getting supposedly. It does worse in most benchmarks
41.
by Bolwin 3mo ago
Claude is very cache friendly, however there have been some inconsistencies with non anthropic endpoints that led to cache breakages
42.
by Bolwin 3mo ago
Glaciers have never been accessible to most people. The night sky has, until recently.
43.
by Bolwin 3mo ago
Yes Americans can do both, unless their boss dislikes it, but that applies the world over.
44.
by Bolwin 3mo ago
Has he made anything else interesting?
45.
by Bolwin 3mo ago
When I use it for fiction, I generally switch models 2-3 times per response. It's basically normal
46.
by Bolwin 3mo ago
Tools come with a tool description in json schema format, but yes your point stands, it is not enough for opus 4.8 which I've also noticed having tool call issues.
47.
by Bolwin 3mo ago
> Yeah, but the biggest plus for open models is that they can never be taken away. In other words, whatever capabilities they reach (even if there will never be another model), those stay forever. In theory yes, but the average person ca
48.
by Bolwin 4mo ago
> Interleaved reasoning and function calling makes this even more dangerous. A model can call functions during the hidden reasoning phase. The reasoning may be hidden but the tool calls are not, how else would the client execute them
49.
by Bolwin 4mo ago
Hah, I noticed the same thing writing fiction with fable. Most models seem to go into a sort of "storytelling mode" where they forget their PhD level smarts. I had a character who is doing repair on a satellite. Most models would
50.
by Bolwin 4mo ago
For Claude models at least, you can tell to just manually think in the output and it works fine. I do it reguralrly because for creative writing and summarization, they seem to believe they don't need to think at all, and get way worse
51.
by Bolwin 4mo ago
None of them even expose thinking anymore. You're seeing a summary
52.
by Bolwin 4mo ago
It's pretty hard to measure because most context rot comes from related context and the model has to be able to figure which parts are truly relevant, which ones are relevant but stale, which ones to ignore etc. Each relevant thing i
53.
by Bolwin 4mo ago
I don't use Claude Code. I use my own handwritten agent (formerly using Pi) and know every token that goes into it. There are zero memories to confuse it. The system prompt is 200 tokens and completely self consistent. Plus I've f
54.
by Bolwin 4mo ago
I see this said often and find it insane given how many times I find opus models making basic recall mistakes at <100k tokens. Personally I consider < 60k to be the smart zone for opus. This is worse for opus 4.7 and 4.8 cause of the
55.
by Bolwin 4mo ago
What do you mean by custom format? Non-json?
56.
by Bolwin 4mo ago
What is the "house" harness for minimax? They haven't released any
57.
by Bolwin 4mo ago
If this is what we get without editors I want every thing I read to be without editors
58.
by Bolwin 4mo ago
Probably not as fast as a simple regex but static embedding models can get stupid fast e.g https://www.flowercomputer.com/news/fast-static-embedding/
59.
by Bolwin 4mo ago
It is a little ironic coming from the most prolific AI poster here
60.
by Bolwin 4mo ago
Also is this an official Microsoft dev blog? Probably not a good look back at publishing hq

More ›