eterm

6 ms·

61.
by eterm 2mo ago
> I just don't accept the absolutist idea that if anything can be gotten around by any means, that means it can't still be sensible policy Now on that I agree strongly, there is a crowd on here that believe that technical weakn
62.
by eterm 2mo ago
I actually agree with you, but you have an optimism about the ability for open source maintainers to have achievable political aims, in a way that I haven't seen happen for decades. What your happier future looks like requires both a t
63.
by eterm 2mo ago
Sometimes when you see your opponents getting out a ramp and a barrel of lube, you can call it a slippery slope.
64.
by eterm 2mo ago
> The breakdown in the pipeline seems to be reliable local diarization Yep, diarization just hasn't been well solved yet. As soon as it has, the quality in note-takers, meeting transcripts, etc, will sky-rocket across the board.
65.
by eterm 2mo ago
AI has clearly been extensively used to improve models. The evidence for this is how far Anthropic went in nerfing Fable to prevent it being used to improve models.
66.
by eterm 2mo ago
From an article someone posted downthread ( https://news.ycombinator.com/item?id=49246228 ) : > Mars could appeal to a more exciting chocolate eater, and a smaller lighter bar could increase consumption per capita: “Whil
67.
by eterm 2mo ago
C# is an interesting case because it shares a common runtime with F#, and F# guarantees TCO in most circumstances ( try / catch can stop it ) . There is a "tail" prefix in the intermediate language (IL) bytecode that F# uses
68.
How to Lie with Benchmarks(blog.eterm.uk)
3 points by eterm 2mo ago|0 comments
69.
by eterm 2mo ago
They're amazing at it, provided you keep asking the right questions. Trust me when I say that in the hands of someone who doesn't have your experience, the LLMs would not be getting the results you get. You might think what you&#x
70.
by eterm 2mo ago
I've found myself with this problem, as I've used LLMs to accelerate solving problems I've wanted to solve, only to find myself unable to share them back, because LLM usage has "polluted" my effort in the eyes of ma
71.
by eterm 2mo ago
That's actually neat, I can finally do a full sweep of something I found after the recent post about Lobsters migration, I came across a query that evaluates differently in MariaDB and SQLite. SELECT CASE WHEN 1 / 5 * 100.0
72.
by eterm 2mo ago
Duckdb CLI is a powerhouse, it can load files as diverse as gzipped json lines, so you can stuff compressed logs straight into a directory yet still easily query them with SQL when you need to.
73.
by eterm 2mo ago
And the official auth URLs are always such random crap that I'm surprised people don't get phished left right and centre.
74.
by eterm 2mo ago
I call this genre of game, "Illusion of choice" games, but to be fair to cyberpunk 2077 it still does it better than any other games of its type. If you think of it as a tree and each decision point branching out to millions of po
75.
by eterm 2mo ago
No-one mourned when that one went to the Google Graveyard.
76.
by eterm 2mo ago
The CSS animation of the encrypted message overwriting the plain is utterly cool.
77.
by eterm 2mo ago
Indeed it makes no sense at all, because Windows 2000 even came out before Windows ME.
78.
by eterm 2mo ago
I've got so many projects I'm simultaneously working on, but the one I really want to get over the line is "Todo driven development" - An edit to the gleam test-runner gleeunit that adds "todo" as a separate re
79.
by eterm 2mo ago
A big shift I've noticed in the past few years is how much Windows is increasingly a second-class citizen for developer focused things. Increasingly you'll see a new tool or library or language, and there'll be installation i
80.
by eterm 2mo ago
Right, but the new phenomena is that we've lost a signal. Used to be if a game looked that good it probably had time spent on the game part too.
81.
by eterm 2mo ago
It's a mix, because some stuff tends to just use `T`, but there's better descriptors elsewhere. There's IList<T> but Task<TResult> There's Action<T1, T2, T3, T4, T5, T6> but also Dictionary<TKey, TVa
82.
by eterm 2mo ago
The expenses tracking revenue does rather paint a picture. No individuals should be donating to wikipedia, it is clear that any donation will be spent as rapidly and as wastefully as possible.
83.
by eterm 2mo ago
Are we saying that it's now a problem that we're not getting scraped? This appears to be a new generation of "SEO", marking itself as a service for getting into AI results? This is not the future I want to be a part of
84.
by eterm 2mo ago
The benchmarking seems a little bit suspect. It links to the bench.cy, but nothing explains how the others were measured. It seems that they're including some start-up time in some that dominates, but not explaining whether it's r
85.
by eterm 2mo ago
A few days ago Claude swore blind to me, "I'm approaching my context window, so you should do...". It was at ~300k / 1M. Very strange behaviour to observe.
86.
by eterm 2mo ago
Fable was good at instruction following but opus 5 seems awful for taking shortcuts to achieve a goal. From a conversation yesterday: > You're right, and it's the same miss again. I wrote the three memory tiers into DESIGN.md —
87.
by eterm 2mo ago
In 2019 that'd be impressive.
88.
by eterm 2mo ago
My preference for visualising disributions is violin plots: https://en.wikipedia.org/wiki/Violin_plot Similar to the ridgeline plot, you can see the distribution better without having to interpret derivatives as you do
89.
by eterm 2mo ago
I think it's kinda cool, it's Gibsonesque, and that usage is old enough now it's rolled from awkwardly dated back to cool, especially now that people no longer tend to talk about "cybersex" and many of the other use
90.
by eterm 2mo ago
I'm not sure why I'm being downvoted, in November last year, the regular teams tier did not let you use claude code, premium was required. They've changed it since, but that's why I said "previously".

More ›