English
Willy
217 posts
Alongside @0xClandestine we got it from 26 to ~30-32 tok/s 🚀
Will cleanup and merge this afternoon
Prince Canuma@Prince_Canuma
DeepSeek-v4 now runs at ~23-26 tok/s on MLX! I made some custom kernels for the sinkhorn and it took gen speeds for 17 -> 26 tok/s. The weights are also significantly smaller thanks to @pcuenq tip about keeping the experts in MXFP4! Now you can use it to power your local coding agents (PI, Open code, Hermes agent or even CC) PR: github.com/ml-explore/mlx…
English
Told you don’t mind the speed 😎
Let me continue cooking!
Prince Canuma@Prince_Canuma
DeepSeek-V4 running on M3 Ultra 🚀 Don't mind the speed, that's gonna improve soon.
English
AI FOOTBALL ANALYSIS. A FULL COMPUTER VISION SYSTEM. BUILT ON YOLO, OPENCV, AND PYTHON.
You upload a regular match video. No sensors, no GPS trackers, just camera footage.
The neural network finds every player, referee, and ball on its own. Every frame, in real time.
KMeans clustering breaks down jersey colors pixel by pixel. The system splits players into teams automatically. Without a single manual hint.
Optical Flow tracks camera movement. Separates it from player movement. Perspective Transformation converts pixels into real meters.
Speed of every player. Distance covered. Ball possession percentage. All calculated automatically.
Four hours of tutorial from zero to a working system.
The model is trained on real Bundesliga matches. Runs on a regular GPU. Python code - take it and run.
Sports analytics is no longer behind closed doors. AI leveled the playing field.
zostaff@zostaff
English
To był ostatni film jaki widziałem, w którym młody chłopiec nauczył się czegoś od mężczyzny.
W którym kobieta musiała stać się silną postacią, a swoich mocy nie uzyskała za fakt bycia kobietą.
W którym "dobro" musi ZABIĆ zło, zamiast je rozumieć i wykazać się empatią.
W którym bohater nie odpuścił ani na chwilę, cel był jasny, a koszt nie grał roli.
W którym przemoc jest związana nierozerwalnie z instynktem przetrwania.
Nie ma już takiego kina, które inspiruje, skłania do refleksji, daje wzorce i kształtuje mężczyzn. Jest tylko nijakość, ideologiczna breja i spłycanie wszystkiego.
Polski
Former BYD and Huawei engineers made a device that turns any bike electric, reaching 32 km/h ⚡
English
@aa22396584 @rohanpaul_ai @ssankar exactly. Having new reaponsibilities of being able to explain your work to other departments is just ridiculous. You are paid the same, but you have like x6 complexity because of like 2 new responsibilities. no human brain can handle that.
English
The optimistic framing here is that AI gives frontline workers direct access to capabilities previously gated behind layers of coordination. The less optimistic reading: cutting bureaucracy is easy to say, but a lot of middle management exists to absorb ambiguity and conflict that nobody wants to handle directly. AI doesn't dissolve those tensions — it just changes who holds them.
English
Palantir CTO @ssankar : "I think AI is going to be the antidote to the managerial revolution of the 20th century.
All this power that was sucked away from the frontline worker, that's being reversed because all the bureaucracy is getting cut".
Rohan Paul@rohanpaul_ai
Marc Andreessen: AI will weaken the manager class, help innovators beat dull managerial systems & force big incumbent firms to innovate fast or collapse. "The innovators need to figure out how to leverage AI to actually do this."
English
LLM Knowledge Bases
Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So:
Data ingest:
I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them.
IDE:
I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides).
Q&A:
Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale.
Output:
Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base.
Linting:
I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into.
Extra tools:
I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries.
Further explorations:
As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows.
TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.
English
@burkov thats quite interesting actually. because it does adapt its sensory input to its sensory output (close mouth and vary the pressure of squeezing). So there should be some planning & guessing ie feeling involved.🤔 I wonder if it can be mapped to human experiences of feeling.
English
@Dan_Jeffries1 Germans are not typically pessimistic, and the article isn’t either. We are somewhat more cautious and critical. The current American culture of manic optimism and dealmaking is more dangerous than even pessimistic assumptions.
English
LiteLLM HAS BEEN COMPROMISED, DO NOT UPDATE. We just discovered that LiteLLM pypi release 1.82.8. It has been compromised, it contains litellm_init.pth with base64 encoded instructions to send all the credentials it can find to remote server + self-replicate. link below
English
AI agents are about to do to finance departments what coding copilots did to software engineering.
An $8B fintech processing $235 billion annually is building autonomous agents for procurement, expenses, and treasury. One client already collapsed 20+ banks into one platform.
Jack Zhang@awxjack
We saved our customers over $1.3 Billion in 2025 alone. That value has helped @Airwallex reach $1.2 Billion in ARR, growing 85% YoY. @Deel, @McLarenF1 , @boltapp and 200,000+ other customers trust us because legacy banking wasn't meant for global businesses: • Opening a bank account in a new country takes weeks • SWIFT transfers take 3-5 days • Other platforms convert your money even when you don't want to But with Airwallex you can: 1. Open an account and get paid like a local in 70 countries Most platforms force you to convert your money into your currency and charge you a conversion fee to do it. With Airwallex, your UK client pays you in GBP and it sits in your GBP balance. Your Australian client pays in AUD and it sits in your AUD balance. When you need to pay a UK vendor or run Australian payroll, you can simply pay from the same currency in your Airwallex account which leads to zero conversion fees. 2. Send and receive money on the same day SWIFT takes 3–5 days and hits you with unpredictable fees on every transfer. But over 90% of Airwallex transactions happen on the same day. Since Airwallex uses local rails to move your money, it also happens at near-zero cost. 3. Issue multi-currency cards instantly Airwallex helps you issue multi-currency cards to your employees across the entire world. And every transaction is automatically synced to your accounting system in real-time. 4. Integrate Airwallex in your product SaaS platforms and marketplaces can also use our APIs to offer these financial services to their customers. In fact, many companies are doing it already. But this is just a glimpse of what Airwallex can do. We’re building the all-in-one financial stack your company will ever need. If you're doing $50M+ in revenue, you could save up to $500k in fees. And that's money back into your business. Sign up for a demo here: airwallex.com/offer/airwalle…
English
@johnrushx @Mshilohgoldberg @en_Clement dude. you are projecting your own ignorance onto LeCun. It looks seriously troubling. Are you mentally developed?
English
I dont share Yann LeCun's point of view
If it's true, how can he explain the intelligence in blind humans?
They have never seen even a pixel in their life, but they are often as smart as a non-blind person.
He makes a typical mistake scientists/devs make by being ego-stubborn
Rohan Paul@rohanpaul_ai
Yann LeCun (@ylecun ) explains why LLMs are so limited in terms of real-world intelligence. Says the biggest LLM is trained on about 30 trillion words, which is roughly 10 to the power 14 bytes of text. That sounds huge, but a 4 year old who has been awake about 16,000 hours has also taken in about 10 to the power 14 bytes through the eyes alone. So a small child has already seen as much raw data as the largest LLM has read. But the child’s data is visual, continuous, noisy, and tied to actions: gravity, objects falling, hands grabbing, people moving, cause and effect. From this, the child builds an internal “world model” and intuitive physics, and can learn new tasks like loading a dishwasher from a handful of demonstrations. LLMs only see disconnected text and are trained just to predict the next token. So they get very good at symbol patterns, exams, and code, but they lack grounded physical understanding, real common sense, and efficient learning from a few messy real-world experiences. --- From 'Pioneer Works' YT channel (link in comment)
English
@johnrushx because touch and balance are the most bandwidth rich senses of a human. Study some neuroscience or something. You with Chollet look laughable to a trained eye
English
@OneInfiniteNow @fchollet it is a highly lossy compression. You have no idea what life 300 years ago looked it. You lost about 99.99999% of that. Text to the most part is useless.
English
Your compression numbers kill the quantity argument. But they might undersell what LLMs are actually training on.
Text isn't raw information. It's the output of the most powerful compression process in nature: millions of humans across thousands of generations, keeping only what survived testing, debate, and selection. LLMs train on that. The bit count undersells it. And JEPA's push toward visual data overlooks it. Text is already the more compressed modality.
What text captures is what survived. Not the process that produced survivors. The loop: hypothesize, test, fail, share, build on failures. LLMs inherited the conclusions. Not the collaborative discovery protocol.
For agents, the practical path might be recreating that protocol. Each agent explores a frontier, compresses what it finds, shares for others to build on. Same loop that produced the training data, just in silicon.
Curious whether ARC-AGI eventually needs to test for this. Collaborative reasoning from shared compressed priors, not just individual program synthesis.
English
I keep reading this take (below) every few months, presented as if extremely profound, and it is just offensively dumb. It confuses data and information, it ignores the fact that not all information is equally valuable, and it ignores the importance of retention rate.
As a thought experiment: if this were true, if your retina cell count were 10x greater, you'd be "trained on 10x more tokens" and therefore you'd be way smarter. Same if their firing frequency were 10x greater. With 10x more retina cells firing 10x faster you'd be "trained on 100x more tokens"!
Obviously this makes no sense -- the signal coming from these cells is extremely correlated over space and time, so their raw information content (what remains post-compression) is extremely low compared to the "raw bit" encoding. The human visual system actually processes 40 to 50 bits per second after spatial compression. Much, much less if you add temporal compression over a long time horizon.
Latest LLMs get access to approximately 3 to 4 orders of magnitude of information more than a human by age 20 (post compression in both cases). About O(10T) bits vs O(10-100B) bits.
And that's just *raw information* but of course not all information is equal, otherwise we wouldn't be spending tens of billions of dollars on training data annotation and generation. Plus, that's only *information intake* but of course humans have far lower retention than LLMs (by 3-4 OOM). You could write a short essay about how incredibly off the mark this take is.
English
