Leynos

6 ms·

1.
by Leynos 21d ago
The latest snapshot of 5.6 Sol feels disturbingly 4o like at times on high in ChatGPT. Although it swears like a sailor
2.
by Leynos 22d ago
It's mostly a porn site these days
3.
by Leynos 27d ago
If you tell people who need digital text from books that they need to destroy books after scanning them, they're going to use destructive scanning and destroy the books.
4.
by Leynos 29d ago
To give you a sense of the computation needed, a spatiotemporal physics simulation of a minimal (493 gene) bacterium's full 105 minute cellular cycle took 4-6 days of GPU hours on two A100s. https://www.sciencedirect.com&#x2
5.
by Leynos 1mo ago
This is why the GDPR (and to a lesser extent the CCPA) is a good thing. The data was supplied for a specific purpose. The handler of the data should have to obtain further consent if they wish to use it for another purpose.
6.
by Leynos 1mo ago
For anyone paywalled, it's on archive.today
7.
by Leynos 1mo ago
Opus 5 tells me no all the time (code cli and web). It's reasons are usually pretty well argued though. Opus 4.7 would flat out refuse to follow instructions to the point where it was just too frustrating to use. I've had refusals
8.
by Leynos 2mo ago
Is this using the server side compaction of the responses API?
9.
by Leynos 2mo ago
1st line of defence, use something like ponytail to enforce brevity. Use property testing and behavioural testing on top of unit testing. Enforce readability standards so you will be able to understand the tests. 2nd line, code review. Do t
10.
by Leynos 2mo ago
If you ask Fable or 5.6 Sol to improve performance, it will generally know to build a benchmark and create a test corpus. I'm not sure where the contrary suggestion is coming from.
11.
by Leynos 2mo ago
The problem with Empire of AI is that it takes such a scattergun approach and doesn't really build a coherent thesis. It is also very difficult to draw clear directional information from the book. I can strongly agree that OpenAI shoul
12.
by Leynos 2mo ago
Zerover for life
13.
by Leynos 2mo ago
I'd prefer a tag to the mounds of "this looks like it is AI generated, I can tell from the pixels and from having seen quite a few AIs in my time" comments. That way the people who reject AI content can filter it out rather t
14.
by Leynos 2mo ago
You'd need the whole edit tree along with all the prompts used along the way, which most people are not yet set up to capture.
15.
by Leynos 2mo ago
For normal building work I use Opus to plan and GPT 5.6 Terra to build. The point is, these are not normal constrained building tasks. Perhaps I should have had more faith in Opus's ability to complete tasks like these, but it just has
16.
by Leynos 2mo ago
The sort of thing Fable and Sol excel at are long horizon tasks. The sort of thing I have been using them for is migrating large numbers of repositories to new tooling simultaneously (adopting new linters, enabling dependabot automerge, rol
17.
by Leynos 2mo ago
Things that I reckon will become a lot more important from a developer's perspective over the next year: - Shaping work so it is more decomposable, legible, verifiable and understandable - Property testing, formal verification (exhaust
18.
by Leynos 2mo ago
Yeah, I went through a period after 4.7 launched of not using Claude for code work at all because of the condescending refusals. (Kept using it for planning and design work). Still had three pretty bad refusals from 4.8, but not the same qu
19.
by Leynos 2mo ago
While there are some coding focused models (composer, for example), the majority of frontier models are pitched as general purpose. The coding harnesses for Claude and GPT are even being repurposed as general purpose knowledge work harnesse
20.
by Leynos 2mo ago
From a purely utilitarian standpoint, direct to cell feels like a good thing to me. Large swathes of Scotland don't even have sufficient mobile connection to send a text message (some people will tell you that's a good thing, but
21.
by Leynos 2mo ago
The prompted response is far from the finished piece of writing. You'd probably want to share the full edit tree and include subsequent refinement prompts in the commit messages.
22.
by Leynos 2mo ago
Yes, that was the point. It made unsafe behaviour visible in a way that could be addressed. I hadn't heard any reports of it being dysfunctional.
23.
by Leynos 2mo ago
As a reader (not as someone who is posting the articles), the AI prose generally doesn't bother me. I'm usually more concerned about what the article says than how it says it.
24.
by Leynos 3mo ago
Samsung are back up 5% today on news of a planned buyback.
25.
by Leynos 3mo ago
Alerts on test fixtures, so suspect it is doing nothing new.
26.
by Leynos 3mo ago
You can see the validation approach they used here: https://github.com/adamraudonis/prylint/blob/main/harness/ch...
27.
by Leynos 3mo ago
Currently, there are things pylint does that ruff doesn't. To use these, I was running pylint on pypy to get it running at a reasonable speed. Having pylint reimplemented in Rust seems like a very useful thing to have from my perspecti
28.
by Leynos 3mo ago
I said "outside of situations where it is required by contract", which I believe would include a CLA.
29.
by Leynos 3mo ago
Which model was used for the benchmark results shown on your GitHub README.md?
30.
by Leynos 3mo ago
Context: https://www.businessinsider.com/what-is-le-chaton-fat-mistra...

More ›