I collect GPUs and bad ideas. Currently serving GLM 5.3 Flash and Deepseek V4.1 Flash locally
- Tweets 18,082
- Following 417
- Followers 2,823
- Likes 8,275
Pinned Tweet
I just invented a new term: Schrödinger's slop.
It isn't slop if you don't look at the code.
Retweet it and let's get this into Websters!
1
9
634
In all my testing with Intel vs AMD for AI servers, I have come with the following conclusion.
Intel can only reach about 1.51us latency and about 44Gb/s p2p bandwidth.
AMD can reach as low as 0.91us latency and around 56.4Gb/s p2p bandwidth.
This greatly affects inference.
1
6
426
Your smart TV is an ad delivery device you paid for. Developers are pointing AI coding agents at their own sets to strip the ads and tracking out. One of the best use case for agents yet.
tech.yahoo.com/ai/meta-ai/ar…
1
1
122
Been getting some great performance out of Deepseek V4.1 Flash (1017) on 4x RTX 6000 Pros (Workstation Power Limited).
6
41
2,321
Meta's CRAM keeps compressed memory in RAM and treats it as RAM instead of faking a swap device. Read-only: 452x faster than ZRAM. With 20% writes: still 5.4x. Not in the kernel yet, but every zram distro and the Steam Deck should be watching. tomshardware.com/software/li…
3
7
316
Man Sold Everything He Owned and Traveled 2,000 KM to Meet His Dream Woman – She Was AI
gadgetreview.com/man-sold-ev…
90
If you are building a high end inference box with gpus like the Nvidia RTX 6000, do not go Intel. The latency is 50% higher than AMD and pcie throughput is 27%+ slower. I have yet to find a fix.
6
2
72
4,294
I wonder if GLM 5.5 Flash can keep up with Deepseek 4.1 Flash 1017 release. Should be interesting, I'm here for it.
5
28
1,915
DHS orders DeepSeek removed from U.S. federal civilian IT systems.
And it goes beyond the chatbot. The ban includes third-party products and services built using DeepSeek implementations.
This is NOT a ban on private use. Yet.
Official DHS order (PDF):
public-inspection.federalreg…
5
2
12
7,402
Websites are starting to close their doors to AI agents. The web was built for humans and polite crawlers; agents that click, book and buy are a different species. More bot walls incoming. theverge.com/ai-artificial-i…
1
2
181
Any time I see a picture of @thsottiaux I picture a monk. Between the hair and clothes.
1
1
176
China proposes export controls on 23 AI technologies — models, agents, clusters. The carve-out is telling: anything already open-sourced is exempt. teddit.net/r/LocalLLaMA/comm…
1
146
I can’t wait to load this
Deepseekv4.1Flash-1017 with lossless csv compression Results are in. The DeepSeek-V4.1-Flash-1017 lossless CSV ran on the KK image from @YourLocalAILab . WE are doing some great work. (i'm running powercapped cards, so WS will be faster on tp4). (i did the CSV encoding myself locally on my hardware)
1
379
If you haven't heard of Local Inference Lab @YourLocalAILab you are not likely running local models as fast as you could be.
2
4
10
2,999
Another day, another exploit.
I'm tired boss.
🚨 BREAKING: A Broadcom Wi-Fi bug lets anyone in range run code on routers without the password.
Quentin Kaiser showed it on Proximus, Orange Belgium and VOO routers and estimates ~1.5M in Belgium are exposed, none patched.
The bug dates to 2010. Broadcom won't assign a CVE.
1
161
Rotate yo keys
We had a security breach last night related to a large scale distillation attack that impacted 1.4% of OpenCode Console users
We have notified them all this morning via email. If you did not get an email you were not impacted, it was a subset of users who had external providers linked
The incident is ongoing so we cannot share more details at this point but we are working with Anthropic, OpenAI, and Google to remediate
1
3
246
26 years, nobody could explain where the NSA's elliptic-curve seeds came from — the prime suspect being a backdoor. An Anthropic research model cracked it in under 30 minutes and found none. anthropic.com/research/ecdsa…
4
3
2,128