Tweet Disematkan
Ed 🍀
1.1K posts
Ed 🍀
@eddysgr
Software Engineering. 🚧 Distributed Protocols - Wallets - AI | open source.
KLA, CPT, DXB ✈️ Bergabung Temmuz 2017
593 Mengikuti484 Pengikut
Ed 🍀 me-retweet
Vercel got pwn’d. Here’s what I’ve managed to get from my sources:
1. Primary victim here is Vercel. Things like their Linear and GitHub got hit with majority of it
2. Env vars marked as sensitive are safe. Ones NOT marked as sensitive should be rolled out of precaution
3. The method of compromise was likely used to hit multiple companies other than Vercel.
# of customers impacted is believed to be limited, but hard to know just yet. Will share more updates as I get them.
Vercel@vercel
We’ve identified a security incident that involved unauthorized access to certain internal Vercel systems, impacting a limited subset of customers. Please see our security bulletin: vercel.com/kb/bulletin/ve…
English
Ed 🍀 me-retweet
Just implemented Google’s TurboQuant in MLX and the results are wild!
Needle-in-a-haystack using Qwen3.5-35B-A3B across 8.5K, 32.7K, and 64.2K context lengths:
→ 6/6 exact match at every quant level
→ TurboQuant 2.5-bit: 4.9x smaller KV cache
→ TurboQuant 3.5-bit: 3.8x smaller KV cache
The best part: Zero accuracy loss compared to full KV cache.
Google Research@GoogleResearch
Introducing TurboQuant: Our new compression algorithm that reduces LLM key-value cache memory by at least 6x and delivers up to 8x speedup, all with zero accuracy loss, redefining AI efficiency. Read the blog to learn how it achieves these results: goo.gle/4bsq2qI
English
Ed 🍀 me-retweet
Ed 🍀 me-retweet
if your 250lb friend isn’t eating at least 100 big arches a month, something is wrong
sunny madra@sundeep
“If your $500K engineer isn’t burning at least $250K in tokens, something is wrong.”
English
@JorgeCastilloPr Isn't this Kimi 2.5 with some post training on top ? 👀 x.com/yulun_du/statu…
English
All companies say their brand new model is frontier-level and outranks all the others.
How long are we going to believe this?
Cursor@cursor_ai
Composer 2 is now available in Cursor.
English
Ed 🍀 me-retweet
SQLite is a masterclass in a library design. It shows how projects of even huge complexity can be very simple and unobtrusive to use. You can learn a lot from them if you are library author, regardless of programming language.
English
Ed 🍀 me-retweet
there is still no substitute for perfectly understanding every single line of code in your codebase
i fall into the trap of just skimming through ai changes to "just make sure it looks good" all the time, and it makes me lose so much time to not perfectly understand every line
English
Something to keep an eye on it. Let's see when they go beyond Llama models.
Would be cool seeing this applied to state of the art models... Interesting stuff
Arvid Kahl@arvidkahl
So Talaas effectively burned a Llama model into a chip. The quantum leap here is not AI on a chip. It's startup latency and what this means for coding. Right now, an AI call takes "ages" compared to a CPU calculation. Network, inference, network, parsing. Hundreds of milliseconds. Seconds even! Talaas turned an async operation into a sync one. Now, we can theoretically expand our programming languages to have extremely fast NATIVE on-server AI calls. Like sub-millisecond. if ($contract->contains("a sentence where one party is significantly weaker than the other") { $contract->review()} // executed in 0.4ms Inline analysis via AI. Can you imagine how much this will change the face of software?
English
interesting things here :
- the ai agent wrote unit tests validating the vulnerable code. 😭😌
- the dev is committing 400+ commits daily, who's reviewing them this fast? what could go wrong?
AI usage is fun and all, but remember there's no free lunch, something gotta pay!
pashov@pashov
🚨Claude Opus 4.6 wrote vulnerable code, leading to a smart contract exploit with $1.78M loss cbETH asset's price was set to $1.12 instead of ~$2,200. The PRs of the project show commits were co-authored by Claude - Is this the first hack of vibe-coded Solidity code?
English
Ed 🍀 me-retweet
Short musings on "cognitive debt" - I'm seeing this in my own work, where excessive unreviewed AI-generated code leads me to lose a firm mental model of what I've built, which then makes it harder to confidently make future decisions simonwillison.net/2026/Feb/15/co…
English
Talking about direct binary generation from LLMs remind of this paper/musing by Ken Thompson many years ago.
X Freeze@XFreeze
Elon Musk predicts that AI will bypass coding entirely by the end of 2026 - just creates the binary directly AI can create a much more efficient binary than can be done by any compiler So just say, "Create optimized binary for this particular outcome," and you actually bypass even traditional coding Current: Code → Compiler → Binary → Execute Future: Prompt → AI-generated Binary → Execute Grok Code is going to be state-of-the-art in 2–3 months Software development is about to fundamentally change
English
Ed 🍀 me-retweet
I don't know if you've noticed, but there's a wave of mass psychosis rolling through tech Twitter, very similar to what we experienced in spring 2023 and spring 2020 (interesting that the periodicity is exactly 3 years)
But the vibes are much darker now than they were last time
English
Vercel has insane egress cost, hosting large traffic sites quickly becomes impractical over there.
Initial step would be to move static assets on dedicate storage cloud providers, that'd reduce drastically the costs, a step further might be to run on a dedicated VPS.
Vercel is still a great platform for quickly going 0 to 1 though.
Riley Walz@rtwlz
Jmail has crossed 450M pageviews, and our Vercel bill has exploded, even after tons of cache mitigation. Many kind people chipped in to cover the bill, but it’s not sustainable. What are solid, cheaper alternatives?
English