Announcing the super-adapter for @Alibaba_Qwen Qwen3.8-27B 🔥
We are closing this week with a cool experimental model - a bigger LoRA that merges all harness adapters from this week - with a simple trick: concatenate all LoRA adapters into one.
Result: beats the thinking model, with thinking with much less tool calls under thinking mode: 19-48% fewer with thinking on depending on the harness
Accuracy gains:
SWE-Bench Pro (hard 51):
→ Hermes Agent: 45% → 61%
→ OpenCode: 47% → 67%
→ pi: 74% → 74%
Terminal-Bench 2.1
→ +4 to +6 points under each harness
We also share the full recipe, experimental results, open questions, etc in the longly awaited full recipe blogpost: huggingface.co/blog/harness-…
GifCo retweeted
Animated SOG (.asog): a compressed format for 4D Gaussian splat clips.
About 11x smaller than the TRiGS files, full spherical harmonics, sound in the file, streams in the browser.
Open source, MIT.
Try it: explore.arrival.space/space/…
Code: github.com/arrival-space/ani…
GifCo retweeted
Allow me to explain the absolutely crooked, retarded financing behind today's Anthropic IPO. While the company is being reported as now being "worth $1.5 trillion" or more, that's all an obvious deception. Out of the 13.2 billion shares that exist in the company (Class A + Class N common stock), only 980 million shares were offered to the public. Thus, over 92.5% of Anthropic's shares have never even been released into the open market, and no one in public markets bought any of those, which means we don't have any legitimate price discovery of the bulk of the Anthropic shares.
Instead, the price that the public paid for the very limited (artificial scarcity) 7.4% of shares offered for sale was CARRIED OVER and artificially applied to 100% of the company's shares even though about 12.2 billion shares have never been priced in any public market.
If Anthropic released all 100% of its common stock shares to public markets, the stock price would plummet much closer to perhaps $5 / share (that's my guess), not the $83 / share that was announced at the public offering (which is now $97 at end of day).
Because if you are selling something and you create a lot of buzz combined with artificial scarcity, the price of that thing will be bid up to an artificial high. Think Cabbage Patch Dolls. But with the perma-smirk of Dario Amodei.
To then claim that ALL SHARES of Anthropic are therefore worth the same price as the 7.4% of shares offered to the public is nothing but sleight-of-hand hucksterism and deception.
Think of it this way: If Anthropic offered just one million shares to the public, at $1,000 per share, but still kept all its other shares off the market, does that automatically mean Anthropic is suddenly a $13.2 TRILLION dollar company?
Of course not. That would be stupid. But that's exactly how the current offering is being rolled out, just with more shares.
The whored-out financial press, which has pushed Anthropic FOMO with the frenzy of a public masturbator, wants you to think Anthropic is now worth over $1.5 trillion. In reality, if all the shares were offered at real supply/demand pricing, the company is probably worth something closer to $67 billion, by my math. And if you count the fact that the company's revenues can't even begin to cover its massive expenditures that have now reached nearly $900 billion in spending commitments through 2032, the real return on investing in Anthropic could be zero.
Virtually none of the people buying Anthropic right now have any clue whatsoever about anything I've explained in this post, because they are financially illiterate sheeple being fleeced by the founders.
GifCo retweeted
Let me make sure I have this straight:
1: Nvidia’s own employees are opposed to the actions of the Nvidiots on the Nvidia board.
2. Nvidia’s foundries (such as TSMC) are opposed to the actions of the Nvidiots on the board.
3. Nvidia’s largest customers (like SpaceX) are opposed to the actions of the Nvidiots on the board.
4. A big chunk of Nvidia’s current prosperity arises from handshake deals between Jensen Huang and various industry movers/shakers. These deals have no associated contract. They are totally dependent on Jensen Huang, who the Nvidiots have driven out of Nvidia.
Have I got that straight?
My god Apple actually released something useful?
Apple just released Depth Pro 2
Sharp and Temporally Consistent Metric Monocular Depth
HF: huggingface.co/apple/DepthPr…
38
It's mind blowing to me how many people actually think Anthropic is a good company and that they're gonna be looking after humanity.
You can argue they have a good business, argue their models, etc. But you have to be living under a rock to think it's ethical or that they care about anyone except having the most power they can have.
It's degenerate, disgusting, dishonest power hungry company.
GifCo retweeted
Today we’re open-sourcing Inworld's Voice Isolator and SAVA (Semantic-Acoustic Voice Activity) models, and publishing the technical report behind Realtime TTS-2.
1/ Voice Isolator: preserve the speaker’s voice.
It removes background chatter, music, and noise while retaining the laughs, breaths, and character of the original recording. It works at 48 kHz, with an offline model for recordings and a realtime model that streams with 20 ms of lookahead.
2/ SAVA: give people room to finish their thought.
Our Semantic-Acoustic Voice Activity model uses both the sound and meaning of speech to detect when someone is speaking and when their turn is complete. It already runs in our voice products and fits on a single laptop CPU thread. In our tests on real meeting recordings, it detected completed turns in a median of 71 ms, compared with 210 ms for the open-source model we tested against.
3/ TTS-2 technical report: share how we built the voice on the other end.
We explain how we trained TTS-2 to follow instructions about delivery, use the same reference voice across languages, and generate streaming speech. We also cover the work behind Flash’s faster, more efficient serving, with training methods, evaluation results, and lessons from failures during development.
Voice Isolator and SAVA include code and model weights under Apache 2.0.
Explore the releases and try them on your own audio here: inworld.ai/open-source
GifCo retweeted
Meet AVM, a generative in-context Autoregressive View Model.
Given one or more images or videos of a complex 3D scene, AVM streams 2D views as you steer a virtual camera. It reconstructs what the context supports and imagines the rest without building an explicit 3D representation.
Find out more at autoregressive-view-model.gi….
Her first public speech should have been to show love & loyalty towards Jensen’s tenure, instinct, sacrifice & legacy, creating the most valuable company in the world & making the AI revolution possible thanks to his foresight.
She should position herself as humble temporary custodian of his legacy until we get clarity after Jensen’s hearing & various inquiries.
Right now she’s inaudible. Nobody wants to “move forward” that easily after losing the visionary behind such a company.
Nvidia's interim CEO urges employees to leave Jensen Huang behind: 'Let's move forward' trib.al/evIy0P3
I mean we all knew she would never fill Jensen's shoes but my god didn't think she would be this stupid.
Nvidia's interim CEO urges employees to leave Jensen Huang behind: 'Let's move forward' trib.al/evIy0P3
28
Everyone posting the same generic motion graphic slop video from Opus when Astra is working hard doing real work.
Working with @AMD, our models identified up to 14% PPA improvement on flagship GPU IP in less than five months—building on designs AMD’s engineers had already spent years optimizing.
The same approach helped improve Jalapeño, our first custom inference chip, with gains verified on real silicon.
New evaluations also show our models getting better at certain chip design tasks.
We’re sharing these results and how we’re working with @Synopsys to bring our model capabilities and their trusted EDA design and validation tools to more chip designers.
openai.com/index/advancing-c…
19
GifCo retweeted
Still a lot of work to do here but achieved SOTA on temporal coherence today!
2.5x improvement over SOTA in both depth error and jitter!
GifCo retweeted
π4 - 0.5B foundation model for real-time 3D reconstruction and camera pose estimation from uncalibrated RGB images.
- All-in-one in a single run
- No drifting tracks
- Handles massive scenes
cool for consumer drone and action camera mapping
yyfz.github.io/pi4/
It's cute she thinks her and the board have a say in this. 😂
🇺🇸 Nvidia confirms Huang will NOT return as CEO!
Nvidia interim CEO Colette Kress says the board has made its FINAL decision not to reinstate Jensen Huang following his resignation.
“Jensen resigned from his position with NVIDIA, and the board accepted his resignation.” 😢
Writer: Samuel
1
38




























