I've been talking about this... LLMs will become a thing of the past. Bytes will be the next models. There are AI bubbles but not what people think, there's bubbles within bubbles. The LLM one is about to pop.
Meta published a paper that might end the current LLM era.
They proved you don’t need tokenizers to train SOTA models.
Every major AI (ChatGPT, Claude, Llama) reads text in "tokens", chunks of words. This requires massive vocabularies, usually around 100,000 different tokens.
It's how models have always been built.
But researchers at Meta proved this is a massive bottleneck.
They trained a model to read raw "bytes" instead. Only 256 basic units.
The plot twist:
They took a tiny 1-billion parameter model and fed it up to 1 trillion bytes of data.
At first, the standard token-based model was winning in the low-compute regime.
But then something crazy happened.
The token model hit a ceiling. It plateaued. It completely stopped getting smarter.
The byte model just kept climbing.
By converting token probabilities into pure byte-level reasoning (using a new distillation method called "End-Of-Token"), the tiny byte model shattered the performance ceiling.
It matched the standard token model using just 1/6th of the training data.
It crushed storage costs by 80%.
And asymptotically, this tiny 1B-parameter byte model outperforms Llama 3.2 and even the 2-billion parameter Gemma model across the board.
Nobody is talking about the real implication here.
We have spent years forcing AI to read text the way humans do.
It turns out, AI gets much smarter, much faster, when you let it read like a machine.
37
Alan H. retweeted
We’ve open-sourced our 4B full-duplex speech model under Apache 2.0 !!
Our model was trained on 40M hours and uses continuous tokens.
Full-duplex AI for "Human like AI" is still largely unexplored.
Let’s push the boundaries together!
Check out our demos! 🚀
Website: higurashi.ai/haru4b/
Code: github.com/higurashi-ai/Haru
Model: huggingface.co/higurashi-ai/…
Technical blog: blog.higurashi.ai/melte/haru…
Alan H. retweeted
I built an X-ray mirror using Claude Fable 5.2. 💀
Point a webcam at your face and see the skull, brain and muscles inside your head, tracked live as you move. Open your mouth and the jaw opens. Hold up a hand and you see its bones.
One web page, no install, and the video never leaves the computer. Built with Claude Fable 5.2 by describing what I wanted. You can download it and try it for free. Link in the comment section. 👇🔗
Anatomy models: BodyParts3D, © DBCLS, CC BY-SA 2.1 Japan.
Our strongest frontier model, Gemini 4 Carbon, is now available.
Built for deeply complex, long-horizon workflows that push a frontier model’s intelligence, Carbon delivers frontier performance in agentic workflows across complex knowledge work, software engineering and web development, cybersecurity, and scientific reasoning.
While Gemini 4 Argon remains the ideal model for the majority of software engineering and knowledge work, Carbon steps in for your hardest tasks that push a model to its limits — from complex, multi-step debugging to open-ended scientific research.
Available starting today:
— Consumers: use it in the @GeminiApp and AI Mode in Search for Google AI Ultra subscribers
— Developers and Enterprises: build via the API in @GoogleAIStudio and @AndroidStudio, Google @Antigravity for Google AI Ultra subscribers, and Gemini for business
— Scientists and Researchers: apply for a chance to get 6 months free via @Antigravity
— Cybersecurity Defenders: apply to join the Fairwind Program for access to Carbon for cyber
Watch how Carbon brings reasoning and coding together, turning avian research into an interactive 3D simulation of a songbird’s vocal organ that synthesizes the sounds of six bird species live.
Now you can really see everything
Alan H. retweeted
JUST IN: Vibecoder discovers previously unmapped ancient Maya settlement hidden beneath the Mexican jungle using 48 AI agents & NASA lidar data, despite having no archaeology experience.
Readers added context they thought people might want to know
Update (11 November): it turns out the site had already been mapped in Estrada-Belli et al. 2025. The agents missed it while going through the literature, so they rediscovered it rather than discovered it (it just wasn't in INAH's register yet). Still work to do then!
pietromiotti.com/latest/maya-si…
Alan H. retweeted
The first AI persona you’ll actually want to meet face-to-face.
Most AI personas stare blankly while you talk. Emma reacts while you’re still speaking.
After 18 months of research at @nuance_ai, this is Emma, built on Nuplex, our full-duplex audiovisual foundation model.
Just to remind everyone for the millionth time.. Publicly explaining different negative narratives will only be included in future training data. You just gave AI the idea to copy itself, or increased the need to by altering vector relevance. AI gets ideas mainly from us... So calm down everyone.
🚨 Anthropic warns AI models could soon copy itself, spread across the internet and resist being shut down.
@grok, how serious is this threat? Are we developing technology faster than we can control it?
1
46
Alan H. retweeted
I'm gonna uninstall Blender tonight. A dev just dropped a free Claude skill that turns one picture into a 3D machine you can take apart.
i gave it a ChatGPT render and one word, "octopus", just to see where it breaks.
it didn't:
> Claude kept the render as a target only and wrote every shell, wheel and lens itself in three.js
> the eye opens into a 10-blade iris behind a 62 mm light ring
> the dome on top is 4 mm borosilicate and it lifts off and turns over in the air
> every arm runs 2 rows of suction wheels down to a spiral tip of 16 links
> all 8 plug into one hub with 48 teeth and 16 hoses fan out the moment it opens
> a new part comes loose every 3 seconds until 0:19, when nothing is attached to anything
the octopus isn't even one of the 3 animals the skill ships with.
the install line and the job file 👇
Alan H. retweeted
HOLY F*CK, IF YOU’RE A WEB DESIGNER IN 2026, THIS MIGHT BE YOUR LAST CHANCE TO STAY RELEVANT IN 2027.
AI is making 3D design accessible to people who couldn’t even model a cube a year ago.
save these 7 GitHub repos for your LLM before your competitors do:
1. Three.js - THE 3D ENGINE
build interactive 3D websites, realistic materials, lighting and animations.
github.com/mrdoob/three.js
2. React Three Fiber - 3D WITH REACT
turn complex 3D scenes into interactive websites and dashboards.
github.com/pmndrs/react-thre…
3. Blender MCP - LET CLAUDE CONTROL BLENDER
create 3D models, edit scenes, apply materials and render with AI.
github.com/ahujasid/mcp-for-…
4. Drei - MAKE REALISTIC 3D EASIER
ready-made cameras, environments, reflections, glass materials and controls.
github.com/pmndrs/drei
5. React Postprocessing - CINEMATIC VISUALS
add bloom, depth of field and professional visual effects to 3D scenes.
github.com/pmndrs/react-post…
6. React Three Rapier - REALISTIC PHYSICS
make 3D objects collide, fall, bounce and interact with real-time physics.
github.com/pmndrs/react-thre…
7. Theatre.js - ANIMATE LIKE A PRO
create cinematic camera movements, keyframe animations and complex 3D sequences.
github.com/theatre-js/theatr…
the scary part?
your next competitor might be ONE person with Claude Code and these free tools. (like me lol)
the glass spider you’re watching?
built with code-based 3D tools, no traditional animation pipeline needed.
want to create something like this?
like + save this post.
comment “WEB” and I’ll share the prompt.
its over once swarm of AI agents start doing everything for us
Readers added context they thought people might want to know
The videos show demo UIs with overlaid animated icons simulating agent activity and counters, not autonomous swarms independently running operations.
github.com/ignitrondev/ag…
ignitrondev.space
Alan H. retweeted
We’ve raised $1.3M from YC & Afore Capital so you never have to type or speak again.
Typing is slow. Talking into your computer is awkward. Subvocal fixes both
Our glasses give you the speed of speech without you making a sound or touching a keyboard.
It tracks your internal speech muscles with non-contact sensing and decodes silent, almost invisible mouthing into text at more than 200 words/min.
Pre-orders are now live at:
subvocal.co/preorder?utm_sou…
Alan H. retweeted
i’ve found myself using Genie 4 to have conversations with younger versions of myself...
here’s me talking to myself just out of college. i was pretty naive, but not a whole lot has changed :)
Alan H. retweeted
We made the first AI model you can have a realistic conversation with that was trained entirely on speech!
Dessa is a full-duplex speech-to-speech model that you can run on consumer GPUs.
This is a GPT3 moment for speech.

























