dun4ev

[Chubby♨️ @kimmonismus]

Anthropics Deputy chief informations officer: „as bad as things have gotten in the past 4 days, they are going to get a lot worse before they get better.”

I am anything but a doomer. However, the current situation seems serious.

[Unitree @UnitreeRobotics]

Unitree Introducing: H2-X Kung Fu Legend Guardian🤖
Powerful performance to serve and guard you, with a maximum endurance of approximately 8 hours.
Please use the robot safely and in a friendly manner.

[Towaki Takikawa / 瀧川永遠希 @yongyuanxi]

3次元CADデータと２次元CAD（AI生成の配管計装図 / P&ID）を連携してシミュレーション・制御。３次元で対応する配管や設備が見えるとごちゃごちゃした２次元配管計装図もわかりやすくなる・・・

[dun4ev @dun4ev]

@RobertJBye I want to try to switch to claude with codex. What interesting things do you offer?

[dun4ev @dun4ev]

@steipete How did you do it? Tell me, please

[dun4ev @dun4ev]

@anxuanng @claudeai @Blender @threejs It's great to play, but what's the practical benefit? How can businesses apply this?

[OpenAI @OpenAI]

Our latest model advances bring stronger capabilities for building and migrating software in low-resource programming languages such as COBOL, Fortran, and Visual Basic.

These languages power software that banks, governments, and businesses depend on. Progress here means understanding decades of business logic and helping developers modernize those systems while preserving what works.

We’re building these capabilities around the challenges developers and enterprise customers bring us. If your team is working on these systems, tell us which languages and workflows we should tackle next.

https://t.co/2TbnkQixJ1

[Chris @ChrisGPT]

How did they achieve this pareto optimality??

[dun4ev @dun4ev]

MiniMax-M3.1-Flash — a Chinese model from MiniMax (October 19, 2026) designed for agent-based tasks, code generation, and computer use.

Specifications:

Context: 1 million tokens
Price: $0.15 / $0.60 per 1 million (cache $0.015)
Input: text, images, video
Reasoning depth: low → max
Speed ~3× faster than the previous M3
Advantages:

Best score-to-price ratio on Automation-Bench (up to ~62.8% at ~$0.03–0.05 per task)
Cost of agent tasks is approximately 70% lower than M3
Competitive results on code (DeepSWE ~73.7%) and Computer Use
Drawback: high tendency to hallucinate according to independent tests.

Available on OpenCode (https://t.co/mYAyq2jW5J)

[dun4ev @dun4ev]

@MarioNawfal Finally, they've brought back the normal Victoria's Secret

[dun4ev @dun4ev]

@elonmusk X has become the best platform for communication and knowledge in the past 20 years. И это правда @elonmusk!

[dun4ev @dun4ev]

I built Atmosphere with Codex, Three.js, TypeScript and Vite.
Fly through clouds, watch lightning hit trees and hear thunder delayed by distance.
Video: Playwright + HyperFrames + FFmpeg. Hosted on Cloudflare.
Sound on: https://t.co/VMea9s7xsw

[Min Choi @minchoi]

Just 48 hours ago Kling 4.0 opened to everyone.

These are not real🤯

7 wild examples + PROMPTS:

1. UGC character at brunch

PROMPT: Raw vertical 9:16 video recorded on a phone's front camera (not on screen), one continuous 30-second take with no cuts. It looks like an ordinary unedited phone clip, not a film: slightly soft image, a little digital noise and compression, deep focus with the café behind her as sharp as her face, auto-exposure brightening and dimming as people pass the sunny window behind her, no colour grading. The phone sits close and a little low, looking slightly up at her; around 9 s a waiter's arm bumps the table, the phone slides and tilts, and she catches it and props it back up. Sadie, a young woman in her early 20s with long honey-blonde waves, light natural makeup, small gold hoop earrings and a cropped cream knit top, sits at a small marble table by a big sunny window in a busy brunch café. Her sister sits across from her, out of frame except for her arm and a hand with a silver ring. Behind Sadie: other tables of people talking, a counter with a glass pastry case, plants, staff moving about, chalkboards too far away to read. Around 9 s a waiter sets down two plates, a lemon ricotta pancakes and a plate of eggs on toast, plus two iced coffees in plain glasses. Around 15 s she turns the pancake plate toward the camera; around 20 s she cuts a bite, eats it and closes her eyes; around 26 s her sister's hand reaches in with a fork and steals a piece of pancake, and Sadie swats at it, laughing. [Sadie: casual young woman's voice, warm and a bit giggly, recorded on the phone's built-in mic from half a metre away]: "Okay, I promised I'd show you what we get. ... We get the same thing every single time. ... Oh, here it is! Thank you! ... Look at this. Lemon ricotta pancakes. ... Mmm. Okay. Yeah. Still the best. ... Hey! Get your own!" [Her sister, off camera: a young woman's voice, dry and amused]: "That's the tax." [Sadie, laughing]: "She does this every time." Audio as captured by the phone mic in a busy café, not studio clean: her voice close but with the room around it, the hiss and grind of an espresso machine, cups and cutlery clinking, overlapping chatter and laughter from other tables, a door bell jingling, a chair scraping, faint music in the background. No subtitles, no captions, no on-screen text.

[Peter McCrory @PeterMcCrory]

Today, we're publishing two new findings from inside Anthropic: how work and hiring are changing at a company at the frontier of AI adoption, and how Anthropic staff expect AI to change the economy.

https://t.co/ehPoxeJGH2

[Sawyer Merritt @SawyerMerritt]

The fact that Tesla was able to keep the development of the Roadster a secret for so long is pretty impressive.

[Neel Nanda @NeelNanda5]

My top takeaway from assessing Gemini 4: Argon is significantly less monitorable than past models, and IMO this trend is likely to continue, like Sol -> Astra. It can evade a realistic CoT+action monitor to do simple sabotage half the time.

We can reliably maintain control for now, with good CoT monitoring, as it still frequently slips up, but we need to be prepared for when that stops being true in future models, which may be sooner rather than later. Among other things, we really need interpretability methods that can reliably capture unverbalized motivations in models, with minimal hallucinations

I suspect this is driven by the model being highly capable without CoT, being capable of 5.5 arithmetic steps in a single forward pass (4 for Fable, 7 for Astra). It seems to omit information from its CoT more often, though in non-adversarial settings the CoT is still very helpful in practice, and remains in legible English. It still needs CoT to do the hardest tasks, but any given task becomes "easier" with each new frontier model.

It is also highly situationally aware, and shows awareness of the existence of CoT monitors and that it may want to evade them, eg it sometimes decides to avoid terms like "scheming" in the output or to keep its thoughts brief as a result. For now, these concealment strategies are pretty obvious in the CoT, but this suggests future misaligned models may successfully evade CoT monitors.

[Боярин Щавель из Ингрии @Abu_Talib78]

Музыкальная пауза в рабочий полдень...

[Tesla @Tesla]

Real life, baby

[Tesla @Tesla]

Do a kickflip!