est

5 ms·

91.
by est 4mo ago
Richard talks about AI been either novel or good Then it shifts to discovery. These seems related but not exactly the same thing.
92.
by est 4mo ago
More realisticly, I hope qwen 3.6 27B on taalas.
93.
by est 4mo ago
That's insanely cool. Now how about actually compress text with LLM that summarize long texts with the non-Euclidean visulization...
94.
by est 4mo ago
Because being critical of AI is the second most boring activity on HN.
95.
by est 4mo ago
This article reads like how to train a LLM without a large corpus your pretrain is doomed to fail Your post-train tricks hardly pays off if your base model doesn't scale.
96.
by est 4mo ago
> render on the terminal like a game engine All those CPU to render this effect https://x.com/cyrilXBT/status/2060617507615207904
97.
by est 4mo ago
> exactly zero software breakthroughs since vibe coding started, other than vibe coding itself Generative AI is meant to be a mimic - Richard Sutton https://x.com/RichardSSutton/status/2061216087744946656
98.
by est 4mo ago
> just plain HTML and some basic CSS Or even better. XML + XLST. True separation of representation and data. Is thousands of nested <div> really a good idea?
99.
by est 4mo ago
MCP is based on a lie: Machines are good at read/generate machine-parsable procotols. Turns LLMs are shit with JSON. Especially those JSON str embeded inside another JSON key-value pairs. Why do smart ppl design a schema like escape JS
100.
by est 4mo ago
to clarify, probably worthless to AI vendors, but might be useful for third-parties.
101.
by est 4mo ago
> I'm going to feed all of my business's data to it Your business data is probably worthless, even considered harmful for the pretrain corpus. Your interactions and decision making process are most valuable parts of the whole b
102.
by est 4mo ago
There are basically two kinds of people in the world, ones that create stuff, and ones that destroys stuff. Defense is a toally different game, and requires a complete new mindset than creativity. Security is something that you miss one the
103.
by est 4mo ago
sounds like a Monty Python sketch ...
104.
by est 4mo ago
> The founding tenet of AI, “saving us” from all sorts of difficult things: climate, disease, poverty, conflict is falling, fast No those are just marketing slogans. The founding tenet of AI is to best match next token according to a rew
105.
by est 5mo ago
I am almost certain this layout is generated by AI, because I vibe coded the exact same newspaper-like style weeks ago.
106.
Show HN: Chatroom with curl command (requires IPv6)(chat.est.im)
2 points by est 5mo ago|0 comments
107.
by est 5mo ago
Your DNS config 5-7 rows are the culprit. Don't point a wildcard domain to Github. It's a wildcard and dangerous.
108.
by est 5mo ago
clickbait.
109.
by est 5mo ago
Why can't you use a web page instead ?
110.
by est 5mo ago
> Senior developers care a lot about stability I saw this yesterday https://trinkle23897.github.io/learning-beyond-gradients/ They are very remotely related yet somehow very close.
111.
by est 5mo ago
Some tricks: docker run -it --rm --pid=container:myapp --network=container:myapp --cap-add SYS_PTRACE nicolaka/netshoot kubectl debug -it pod-name --image=nicolaka/netshoot
112.
by est 5mo ago
> Even a careful minimal build with uv lands at 282MB. wtf. Is there a cleanup tutorial somewhere?
113.
by est 5mo ago
WebTransport requires some speicific server setup. cldouflare doesn't support WebTransport well.
114.
by est 5mo ago
I have a question, does LLM follow these MANDATORY or DO NOT SKIP during pre-train, like how people write a comment paragraph on reddit corpus, or is it just some post-train alignment habbit?
115.
by est 5mo ago
I think blogs should have a universal comment function, I'd love to engage ith the author but many blogs don't bother providing a basic textarea.
116.
by est 5mo ago
> doesn’t quite work for cases In that case, the "Co-authored-by: Copilot" method doesn't quite work either. You have to split the commit somehow. > this is who’s driving the AI As indicated by the consistent user.email
117.
by est 5mo ago
I am using a different approach. `user.email` is always my email. `user.name` is either my account name, or model name like `gpt-5.5-high`. I can easily filter & blame which line was written by me or some specific AI
118.
by est 5mo ago
Today junior assembly language programmer are all gone, too.
119.
by est 5mo ago
I added these AmbientCapabilities=CAP_NET_BIND_SERVICE CapabilityBoundingSet=CAP_NET_BIND_SERVICE NoNewPrivileges=yes to my .service. Is it good enough?
120.
by est 5mo ago
the go-to method is using frames.

More ›