Co-Founder BEHOLDER | Ex: Blizzard, Sony, Obsidian / Designer / Artist / Misfit / The Universe prefers novelty.
Austin, TX
Joined April 2008
- Tweets11.7K
- Following2.7K
- Followers1.4K
- Likes131K
Ian retweeted
Tim Sweeney is criticizing Valve again over Steam’s requirement for developers to disclose their use of generative AI.
The Epic Games CEO called the policy “abhorrent,” arguing that developers are being forced to make a “political statement” that could be used against them by a “hate mob.”
Sweeney has been criticizing the policy since 2025, previously comparing the disclosure to a “Scarlet Letter” and arguing that AI will eventually be part of almost every game production process.
Ian retweeted
new from us at @blocks
Meet Dessa, our AI model trained from scratch on nothing but speech.
We believe speech is the natural interface for technology. So we built Dessa.
And we’re inviting AI safety researchers to explore Dessa.
Learn more: block.xyz/dessa
Ian retweeted
We're excited to announce abliterated-model-large-v2 (GLM 5.3 abliterated) is live on @AskVenice!
Abliterated Large V2 by @abliteration_ai is now on Venice.
If your job is breaking things before attackers do, "I can't help with that" isn't an option.
Exploit analysis, payload review, threat modeling, red-team scenarios. No refusals, no logs tied to you.
Available anonymously.
everyone is about to Schiff their pants.
🚨📉 Gold has dropped below $4,000/oz for the first time since July, amid a broader selloff...
Investors sold off gold to cover margin calls and capital requirements during a deep equity and AI stock downturn.
Analysts note that key technical support sits near the $3,900/oz mark, but the dip below $4k still hurts...
Source: TradingView / Writer: Michael
Ian retweeted
Clearly, I will not be getting sleep tonight
huggingface.co/MiniMaxAI/Min…
Ian retweeted
Claude Code is helping build a cyberpunk pixel-art game where even the environments are generated through code.
City of Animals mixes Monkey Island-style adventure with fast combat inspired by Huntdown and Replaced.
The workflow is pretty interesting:
> Unity + C# for the game
> Claude Code writes most of the code
> characters and bosses are drawn by hand
> PixelLab creates animation drafts
> Aseprite is used for frame-by-frame cleanup
> Claude writes Python scripts that draw most backgrounds and props
Claude is also kept on track between sessions using a `CLAUDE.md` file for rules and a `STATUS.md` file for the current to-do list.
Around 785 commits after just a month of development.
Ian retweeted
Vibe coding with Unreal Engine for the first time!
Tripo CLI → Blender → Unreal Engine
The animations and VFX are on a whole different level.
The only downside? I can't deploy it directly to the web. I'll need to find somewhere to host the game so people can play it.
Ian retweeted
NVIDIA sold millions of 3090s with 50% of the tensor speed switched off. Not missing. Switched off.
1 kernel module flips it to full, and my BF16 tensor pipe went from 82 to 155 TFLOPS. Nothing else changed on the cards, and the outputs match the stock run byte for byte.
All so the gaming card wouldnt eat into workstation sales. Every 3090 owner ran at 50% for 6 years, so a pro card could look worth the price. Fine, its their business model.
And LLMs are the least interesting place for this. Decode waits on memory, and my engine runs its experts in int8. Still +10% on prefill.
The models that live on BF16 tensor math are everything else. Qwen Image 2.1 and FLUX 3 Klein, where every diffusion step is big matmuls. MiniMax H3, which takes 4 to 7 minutes on my cards. Music and speech models. And training, every LoRA anyone ever trained on a 3090 went through exactly the path that was cut.
Thats what I measure next. This is only the start.
My 3090s just got 1.9x FASTER at BF16 tensor math!!! Same cards, same driver, same 300W, one Linux kernel module.
bf16 tensor pipe: 82 -> 155 TFLOPS
cuBLAS BF16 GEMM 8192: 64 -> 99 TFLOPS
TF32: 41 -> 79 TFLOPS
NVIDIA ships every 3090 with fp32-accumulate tensor math at half speed. The driver even reports it: FMLA32 = 1/2. gpu-unlock by Zagitta sets it to full before the NVIDIA driver loads. After a cold boot all 4 of my cards say FMLA32 = full.
Real inference got less than I predicted. Strata, Qwen3.8 Flash-Next on 2 cards, prefill:
4K: +12%
10K: +12%
90K: 3205 -> 3536 tok/s, +10%
Before installing I said +14 to 19% at 90K, so I was wrong. The GEMM phases got 23% shorter, but they are only about 40% of Strata prefill, and the int8 experts, copies and CUDA core kernels didnt get any faster.
Outputs did not change. With the same expert cache size, my log-prob dump over 1996 teacher-forced positions is the same file byte for byte as the stock run before the module.
Next is Strata itself, now its everything around the GEMMs that makes prefill slow. How I set it up is in the replies.
Ian retweeted
WTF, $3000 Samsung TV has an always-on mic, automatic content recognition, geo tracking, network mapping, and ships your data to Samsung + 400 partners.
Bro Rooted a brand-new 2025 Samsung TV after noticing absurdly targeted ads.
Firmware was locked down, so zero-days it was.
Got a root shell.
then found: always-on voice service, ACR, geo, full telemetry pipeline, and four separate DRM systems.
This isn’t a bug, It’s the product bro.
i wish i could say this is unique and isolated.
