R&D Programmer @ Xplored https://xitter.gq/t.co/H13kiPe2v1 https://xitter.gq/t.co/CdWWpiOIwh
Genova
Joined June 2009
- Tweets8.7K
- Following1.4K
- Followers554
- Likes23.8K
Pinned Tweet
This is basically my life philosophy.
xitter.gq/fermatslibrary/status/…
Gianluca Alloisio - Teburu.com shop is live! retweeted
Earlier this week Tristan published his top YouTube channels:
thume.ca/2026/11/01/youtube-…
Which looked excellent so clearly I had to turn it into TristanTube:
karpathy.ai/tristan-tube/
47K super nerdy videos +TV mode for laid back watching. (One-shotted with Claude Code in 3 minutes)
Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5:
hackertube.org/
- Excellent recs without any logins (state is kept in local browser session)
- Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads
- Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version
- HackerTube Longs ✊
A bit more on the process:
I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing.
Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.
Gianluca Alloisio - Teburu.com shop is live! retweeted
I've been making claude optimize llama.cpp for rx7900xtx on my setup, that is pcie4 ddr4 with 5900x as CPU, 64GB 3200MHz DDR4. I'm currently making it optimize the prefill and attention kernels with the AMD perf tools. Here is a summary:
Gianluca Alloisio - Teburu.com shop is live! retweeted
Highly recommended paper from NVIDIA.
(bookmark it)
There is a lot of debate about whether to use context compaction aggressively.
They show that with frequent compaction, frontier language models lose little to no accuracy on a broad range of long-horizon and long-input tasks while keeping their context short.
With one action per context, Claude Opus 4.5 scores 70.2 on SWE-bench Verified against 70.0 with the full history. Median peak context drops from 32K to 8K tokens.
The agent carries a short summary of its state between steps instead of the whole transcript. Simple compaction at the end of a conversation often does better than the full history, and Opus goes from 70.0 to 74.8 with it.
This also matters for cost. Later tokens cost more to generate than earlier ones, up to 7x on frontier open-weight models.
The pattern holds on Terminal-Bench, FRAMES, BFCLv3 and multilingual SWE-bench, with modest losses in some settings.
Before paying for million-token agent runs, it's worth testing aggressive compaction on your own tasks.
arxiv.org/abs/2611.05306
Chat with Paper: academy.dair.ai/papers/you-d…
Gianluca Alloisio - Teburu.com shop is live! retweeted
Today we are happy to share one of our first open research releases: Latent Sparse Attention (LSA), a step towards subquadratic attention that matches accuracy of full attention at ultra-long context.
Sparse attention attends to only a few tokens per query, but it has to find the right ones. At 10M tokens, that search takes 98% of the attention layer's time. LSA makes sparse attention up to 22.7× faster and full attention 72.4x faster.
With @kubicek_ales @caciolai @PengLiangzu @SukritSumant @Ben_Hoov @FraserGreenlee @RMBattleday @radim_urban @thoefler @jcrwhittington
Gianluca Alloisio - Teburu.com shop is live! retweeted
Do NOT buy a 512GB Mac Studio ‼️
Two 256GB Mac Studio M5 Ultras are ~$700 cheaper and have ~1.7x faster prefill and decode
Price: $22,299 vs $21,598
Both are the 36-core CPU / 80-core GPU
You can connect 2x 256GB with one Thunderbolt 5 cable + RDMA, which is fast enough to run one model across both
Prefill (compute)
- 1x 512 does ~135 TFLOPS
- 2x 256 does ~215-245 TFLOPS
- Apple claims up to 3x for a 4-Mac cluster, which works out to ~1.7x for two
Decode (bandwidth)
- 1x 512 has 1.2 TB/s
- 2x 256 gets ~1.9-2.2 TB/s effective
- With tensor parallelism each box reads its half of the weights at the same time, which works out to ~1.7x for two
GLM-5.3 goes from ~33-40 tok/s to ~55-70
DeepSeek V4.1 Flash goes from ~45-60 tok/s to ~75-100
Both are 4-bit with spec decode on, estimated from M3 and M5 Ultra benchmarks
If you want Qwen3.8-Flash-Next or GLM-5.3-Flash, a single 256 already runs them well and you don’t need either setup. The 512GB question is about DeepSeek V4.1 Flash (552B) and GLM-5.3 (753B), tied as the best open models on Artificial Analysis v5. They need 310-420GB at 4-bit, so neither fits on one 256
Gianluca Alloisio - Teburu.com shop is live! retweeted
Have been working on various thing including a (hand crafted) engine for the past 7 months with the goal of large forests with highpoly vegetation (around 1M instances * 4M tris in this case). Haven't really been motivated to post on X lately, but I will go into more detail soon.
Gianluca Alloisio - Teburu.com shop is live! retweeted
I've been reviewing a lot of code written by Opus 5.5, and it still seems to make a lot of bad software engineering decisions. The kind that will accumulate/calcify into unmaintainable software over time. IMO even the agents will struggle to maintain vibecoded software.
Gianluca Alloisio - Teburu.com shop is live! retweeted
Writeup: preferencemodel.com/blog/ter…
Of the 65 tasks:
60 tasks had at least one issue
44 could be passed by cheating / reward hacking
41 tasks were unfair, e.g. grading on unstated requirements
325 issues in total
Gianluca Alloisio - Teburu.com shop is live! retweeted
The new Developer Diary for the The Bad Karmas: Curse of Cthulhu - Volume 6 is now online on BoardGameGeek! 🐙
Topic of the Day: the brand-new 2v2 game mode called “Curse of Competition” ⚔️💥
👉️ buff.ly/SOQ8YK3
#teburu #tabletopgaming #boardgames #thebadkarmas
Gianluca Alloisio - Teburu.com shop is live! retweeted
maybe the post referred to sdf modeling directly by codes like what iq did. For general sdf modeling, artists can make brilliant models with intuitive sdf toolset as well.
In fact, I'm pretty sure the only person in the world who can build models like this by hand is @iquilezles. My AI-assisted process borrows (and outright steals) from his work.
In that spirit, I dub the model “Inigo the Dragon”: cs.cmu.edu/~kmcrane/Projects…
Sorry @iquilezles! [3/n]
This would be interesting for gondolin and all the pi-gondolin extensions out there, mine included (unreleased)
"NVX is an ultra-light micro-VM sandbox for running untrusted workloads with hardware-enforced isolation. It is built on top of OpenVMM and runs Linux as a guest."
github.com/microsoft/nvx
It looks like it works with Windows Hypervisor Platform (WHVP/WHP) and Linux MSHV.
Gianluca Alloisio - Teburu.com shop is live! retweeted
People of Pi: We've shipped Pi 1.0 with Pi Durable. Go make them yours.
earendil.com/posts/pi-1-0/
Gianluca Alloisio - Teburu.com shop is live! retweeted
AMD on Windows + MCP Server by the way.
Fixing some issues along way.
github.com/Niko1221/Strata/r…
Gianluca Alloisio - Teburu.com shop is live! retweeted
Run Qwen3.8-Flash-next on 8GB+ AMD GPU 🔴
RX 7900 XTX runs the model at a sustained 52-60 output tokens per second + 1250 prompt processing tokens per second.
+ Enable KV Cache - k8v4 for + 10% boost in performance with no quality loss
github.com/Niko1221/Strata
Gianluca Alloisio - Teburu.com shop is live! retweeted
store.steampowered.com/app/3…
[ 2/2 ]
