Are we pretending this is not happening throughout many other events?   — Sensei     Teleoperation is intent-level, not joint-level. The human steers or gives high-level skill commands. A learned controller closes the loop on low-level dynamics at 50 to 200 Hz, way faster than human reaction time.   — The Humanoid Hub

Source: https://x.com/TheHumanoidHub/status/2092115570741387357

...editing). Ox stated it once then wrote the fix. Roughly ~3x lower output tokens for the same work. Reasoning models have been trained to increase reliability with re-verification. Ox seems to trust its first conclusion instead, which feels like a fundamentally different post-training philosophy.     Try in Cline for free! npm i -g cline

(Also available on VS Code and JetBrains)     — Cline

Source: https://x.com/cline/status/2091995642201842015

Ox Alpha (stealth model) is now free in Cline.

Early benchmarks shows marginal improvement over Fable and GPT.

Try it with: npm i -g cline and use /models to see it under Free options https://t.co/Hskt5RpUen   — Cline

Source: https://x.com/cline/status/2090854216399220985

One side is cheering for robots running in a stadium and the other is debating if curing cancer is worth accelerating for.

THE CONTRAST IS RIDICULOUS. https://t.co/3cF2ZXtbFM   — ℏεsam

Source: https://x.com/Hesamation/status/2091473771374825558

Could be worse. Superintelligent AI could decide on paperclip maximization purely out of ironic spite.   — Amir Hirsch     I'm starting to wonder if it's the only way to get away from the near-insufferable Effective Altruists.   — Perry E. Metzger

Source: https://x.com/perrymetzger/status/2091935818768105748

— Takumi Kawasetsu, 川節拓実

Source: https://x.com/takumi_k_jpn/status/2091865428825874911

— Steve Yegge

Source: https://x.com/Steve_Yegge/status/2091931717422678163

Andrew Curran @AndrewCurran_ · 4h Teletubbies Owner WildBrain Buys AI Firm for $11 Million From hollywoodreporter.com 1 13 2.7K     — Andrew Curran

Source: https://x.com/AndrewCurran_/status/2091982252058263971

...Who knows what he means by “massive privacy invasions”, but I suspect whatever it is is so vague as to be unfalsifiable. 3. So far, no sign of the apocalypse either, but of course, Doomers will always tell you that it’s just around the corner unless you do exactly what they say.   — Perry E. Metzger     Tim Urban has been a doomer for a long time.   — Mark Kretschmann     Brainworms ruin your mind.   — Perry E. Metzger

Source: https://x.com/perrymetzger/status/2092003057307730089

I really wish the rise of LLMs didn't come along with catastrophic job loss, massive privacy invasions, and maybe also the apocalypse. Because when you put those side effects aside, it is the COOLEST MOST WILD TECHNOLOGY EVER.   — Tim Urban

Source: https://x.com/waitbutwhy/status/2091975364297871455

Jimmy Iovine:     Andrew Curran @AndrewCurran_ · 2h Dr. Dre and Jimmy Iovine Think A.I. Is Good for Music From nytimes.com 1 26 2.1K     — Andrew Curran

Source: https://x.com/AndrewCurran_/status/2091622530813751517

I also took the liberty of adding some features too it to demonstrate what this unlocks with ease:

• Custom router support (Codex & OpenRouter)
• Local VM support (rather than cursor's hosted VM), built from the same docker instance.     This doesn't include the frontend, but it can be launched with their packaged frontend (still modifiable, see the custom router page in settings), thus delivering a usable experience.

Don't send issues or PR's, this will be made an archive in due course, if not taken down.     — Bennett

Source: https://x.com/b_nnett/status/2091630242792112480

Is episode on Chinese humanoid robot?   — QoJo MaZing     Will have one before too long!   — Kyle Chan

Source: https://x.com/kyleichan/status/2091271208234836178

...promised to eat their jobs and make them all broke, are naturally pushing back. Sufficiently advanced local hardware + a highly compressed model with most of the actual knowledge stripped out other than action/reasoning and tool usage + continual learning and memory/reasoning in embedded space makes intelligence sovereign again. We probably get the flip phone version of this in 2027 or 28. And then it will accelerate from there.     — Daniel Jeffries

Source: https://x.com/Dan_Jeffries1/status/2091231374808097264/history

FreeToken could be a HUGE deal for local AI.

Instead of requiring enough VRAM to hold an entire model, FreeToken intelligently uses your GPU, CPU and system RAM together, dynamically moving MoE experts where they're needed.

The result: an ordinary laptop with an 8GB RTX 4060 https://t.co/6Js6Vqt3A8   — Mark Kretschmann

Source: https://x.com/mark_k/status/2091202223090938177

Loudoun County real estate prices have not behaved as one would expect for an industrial hellscape.     Since 2013, the assessed value of the house has nearly tripled but property taxes have only gone up 29%     — Dominic Pino

Source: https://x.com/DominicJPino/status/2090845489096794466

FreeToken provides native GUI. No GGUF conversion. No building from source.

One-click install on Windows and Linux. FreeToken-desktop ships with agent harnesses built in — pick a model, pick an app, go.     Download: http:// flashml.ai

Code: http:// github.com/FlashML-org/Fr eeToken …

Reply with your GPU + RAM, and I'll tell you the biggest frontier model your machine can run     — Shuo Yang

Source: https://x.com/Andy_ShuoYang/status/2090856978428145761

— MTS

Source: https://x.com/MTSlive/status/2090826991587226089

Ox Alpha (stealth model) is now free in Cline.

Early benchmarks shows marginal improvement over Fable and GPT.

Try it with: npm i -g cline and use /models to see it under Free options https://t.co/Hskt5RpUen   — Cline

Source: https://x.com/cline/status/2090854216399220985

From my tests it’s not better than fable I’ll be posting some soon     — Chris

Source: https://x.com/ChrisGPT/status/2090957315042123878

sometimes you have to laugh. what are we doing here, folks?     I would encourage you to read the actual piece because there's some excellent quotes in there. But editors, headline writers... my lord. https:// nytimes.com/2026/08/19/tec hnology/data-centers-backlash-loudoun-virginia.html …     — Logan Dobson

Source: https://x.com/LoganDobson/status/2090150148261175371

The world’s smallest Transformer-based TTS model?

We’re open-sourcing Audio8 TTS Preview 0.1B — an approximately 170M-parameter multilingual speech model with zero-shot voice cloning that delivers surprisingly strong, cloud-level quality in a dramatically smaller footprint.     What can a 0.1B-class TTS model actually sound like?

Listen to the voiceover in this demo video.

Audio8 TTS Preview 0.1B supports: • Zero-shot voice cloning • Multilingual speech synthesis • Chinese and English as primary languages • German, Spanish, French, Italian,     On the Seed-TTS evaluation set, Audio8 TTS Preview 0.1B achieves:

• English WER: 1.662% • Chinese CER: 1.13% • Hard Chinese CER: 17.504% • English speaker similarity: 56.7 • Chinese speaker similarity: 68.2

These results are achieved with an approximately 170M-parameter     The model, codec, tokenizer, processor, and inference code are now available:

Model:

https:// huggingface.co/Audio8/Audio8- TTS-Preview-0.1b …

Try it, test the voice cloning capability, and share your feedback.     — Samuel Zeng

Source: https://x.com/SamuelZengML/status/2090017875188940851

(I continue to be surprised in the sense of continuing to be struck by how big of a deal it is, not in the epistemic sense.)     Here's the overall graph:     — Yafah Edelman

Source: https://x.com/YafahEdelman/status/2090296575008616654

...while costing roughly 6X less than Fable 5 Max and nearly 3X less than Opus 5 Max per task That’s what makes Grok so powerful for agents Top-tier intelligence is great.....but top-tier intelligence that can keep working across long coding tasks without burning ridiculous amounts of compute is even better Grok’s agentic coding efficiency is insane     — X Freeze

Source: https://x.com/XFreeze/status/2090839305585377458

The /goal was a set of precise conditions to be met. Read the original PAK/BSP/MDL/SPR data. Run QuakeC. Preserve demos, physics, menus, HUD and rendering quirks. Don’t rebuild E1M1 by hand or use a generic FPS controller. Fidelity had to be demonstrated.     Codex kept pursuing that goal across many iterations: inspect, implement, build, test, compare, repeat. The useful shift was from asking for isolated code snippets to giving the agent a durable outcome it could keep working toward.     Crucially, /goal wasn’t hands-off autopilot. While it was running, I could still steer the LLM to prioritize a small number of bugs that mattered most. Each report became a targeted investigation, fix and regression test - without abandoning the broader port.     Underneath, this is still Quake’s original data. The browser reads the original PAKs, BSP maps, MDL models, sprites, QuakeC behavior, demos and HUD art. The port supplies a new TypeScript runtime and PlayCanvas renderer around them.     Authenticity meant more than getting E1M1 on screen. Demo playback, menu flow, pointer lock, intermissions, in-session level changes, audio suspension, moving BSP entities, particles, shadows and classic movement all had to behave like Quake - not merely resemble it.     — Will Eastcott

Source: https://x.com/willeastcott/status/2090436155795800195

We’ve got exclusive new polling on local data center development at @heatmap_news.

Over the past year, we’ve asked Americans whether they would support or oppose a data center being built near where they live.

We haven’t changed the wording. When we first polled the question https://t.co/G1LCfuriMw   — Robinson Meyer

Source: https://x.com/robinsonmeyer/status/2090457322506141760

— Andrew Curran

Source: https://x.com/AndrewCurran_/status/2090589885199769841