Tony Davis

@lauriewired i hear this music in my nightmares

Telephone Hold music sounds really bad. Mostly because it’s mapping complex instruments to a human throat. Many phone lines these days use CELP, Code-Excited Linear Prediction algorithms. Music breaks down in weird ways when you turn a piano into…speech.

@Sauers_ Sparsity regularization?

Any computation graph can be realized as a neural network. Good compilers optimize computation graphs to eliminate redundancy. We currently have no such compilers for neural networks. Why?

@kalomaze The point is that the models memorize solutions instead of generalizing the process of programming

the kind of person who asks "but does this transfer generalize to Brainfuck?" is simply not being a serious person tbqh

not shocking at all; the models don't want to write in byzantine esoteric languages instead of python or rust or whatever

@tenobrus Exactly! If a small open source model matches frontier performance, the frontier will move. Obviously

i think this will almost certainly happen, but when it happens no serious knowledge worker will use it, because frontier models at the time will be so much more capable using last year's model on a local box would be the equivalent of purposefully giving yourself brain damage

@askalphaxiv check out markxiv.org if you want full text and images

Introducing MCP for arXiv Let your research agents stand on the shoulders of giants Fast multi-turn retrieval, keyword search, and embedding search tools across millions of arXiv papers 🚀

@InferXai Yes highly interested. Do you support custom models?

@TonyD993 We are in Private beta. Happy to give you access if you want to try it out.

A couple of weeks ago we demonstrated 1.5s cold starts for a 32B model. Today we’ve pushed it even further. We’re now seeing sub-second cold starts for models of this size. Why does this matter? Cold start latency is one of the biggest barriers to true serverless inference. If models take 40 seconds or minutes to start, developers are forced to keep GPUs running 24/7. Fast cold starts change the developer experience completely. Models can finally run on demand instead of sitting idle. We’ll be talking about how this works during our live technical webinar on Wednesday, March 18 at 8:30 AM PST. Link in the comments.

@InferXai Not general access yet?

@TonyD993 Yes, we expose an API. It’s compatible with the OpenAI-style interface, so you can plug it into existing workflows pretty easily.

@JustDeezGuy Somethings gotta give somewhere. There's a minimum amount of complexity in a given problem. That can be wrangled any number of ways. Most pure FP or OOP I've read are terribly hard to read and maintain.

@TonyD993 Not necessarily.

Pure FP is about “reducing the state a developer has to hold in their head” to 0.

@LammingLab Isn't leucine implicated in tons of aging effects because of mTOR activation?

Please enjoy the latest publication from our lab "Restriction of Individual Branched-Chain Amino Acids has Distinct Effects on the Development and Progression of Alzheimer's Disease in 3xTg Mice"

@ChShersh Yes 911? this post right here

Here me out. React but in C++

@contextkingceo How's it different from any other graph db?

We've raised $6.5M to kill vector databases. Every system today retrieves context the same way: vector search that stores everything as flat embeddings and returns whatever "feels" closest. Similar, sure. Relevant? Almost never. Embeddings can’t tell a Q3 renewal clause from a Q1 termination notice if the language is close enough. A friend of mine asked his AI about a contract last week, and it returned a detailed, perfectly crafted answer pulled from a completely different client’s file. Once you’re dealing with 10M+ documents, these mix-ups happen all the time. VectorDB accuracy goes to shit. We built @hydra_db for exactly this. HydraDB builds an ontology-first context graph over your data, maps relationships between entities, understands the 'why' behind documents, and tracks how information evolves over time. So when you ask about 'Apple,' it knows you mean the company you're serving as a customer. Not the fruit. Even when a vector DB's similarity score says 0.94. More below ⬇️

@shakoistsLog It is smart for about 3.4 seconds before it literally starts outputting gibberish or chinese. Claude still wins.

i need you be really serious with you guys for a second, no trolling. gpt 5.4 xhigh codex is like 50% smarter than opus 4.6 in claude code. maybe 70%. if you need intelligence make the switch.

@DistStateAndMe @erfan_mhi Lfg

@TonyD993 @erfan_mhi We are working on it

We just released the model + technical report for Covenant-72B. The largest LLM ever pre-trained on a fully decentralized infrastructure. 72B parameters trained over the open internet with permissionless GPUs. This is a big step toward making decentralized pre-training actually practical. Amazing work by @covenant_ai team.

@tplr_ai Any incentives to train?

We just completed the largest decentralised LLM pre-training run in history: Covenant-72B. Permissionless, on Bittensor subnet 3. 72B parameters. ~1.1T tokens. Commodity internet. No centralized cluster. No whitelist. Anyone with GPUs could join or leave freely. 1/n

@CtrlAltDwayne Context rot will kill you eventually

AI generated code will add three null checks, two fallbacks and a try/catch around something that has never once failed in production. And you know what? It still runs. It still passes tests. It shipped in 4 minutes instead of 4 hours. I genuinely do not care about the extra null check.

@mov_axbx This is what I've been doing

@itsolelehmann This will NEVER happen. Banks don't have apis because they rely on browser fingerprints to prevent fraud.

I need a personal bank account with api access to i can do simple banking tasks using my agent