Tensor-Slayer
@TensorSlay
Followers
556
Following
2K
Media
661
Statuses
10K
Exploring Direct Tensor Manipulation in Language Models: A Case Study in Binary-Level Model Enhancement:
1
9
63
Hyperscalers that spent 2024/5 hoarding NVIDIA cards are discovering that a rack of Blackwell GPUs is useless without a 150 MVA step-down transformer, a slab of Japanese ABF substrate and a monthly allocation of indium-phosphide lasers. The marginal cost of a delivered
"The era of cheap silicon is over." Interesting words from the Nothing Phone guy
0
0
0
One question : Is he Chonky ?
20%
0
0
0
Time to repost the banger which reached no one because X algo buried me after calling out Theo and getting blocked
Seems like everyone has solved the king of RL problems i.e credit assignment function of process reward models in their tweets. I, for one, am lost on the counter to the counter-narrative intuitively.
0
0
0
I like this direction. Sandbox agents only to let em run wild.
We just announced Mogra! Mogra gives agents what they actually need: sandboxed execution environments + persistent state across sessions + deployment. More coming! Harness > Model.
1
0
3
Hard disagree with skills. Test this yourself : Generate a basic web page. Then use Claude frontend design skill to generate the same page. The difference is night and day. This is a “low hanging fruit” example btw.
A person's first exposure to an AI agent really does matter. When I look at the Claude Code harness, to me, it looks like the inferior harness. Anthropic are really good at creating all these stuff to augment the model, that ends up not mattering. MCP don't matter, the model
0
0
2
Imagine how it feels for 2010 account holders
X's timeline has gone under a horrendous mode collapse the algo will basically mask you out if you don't echo the trending keyword this is why it feels so slop to most old timers we need to retvrn to the days of high entropy novel bangers
0
0
1
Not sure I understand this direction. CC was supposed to be a new way to interface with code. If we wanted we already had the ability to ….use IDEs to see the diffs. Wondering if this is to cater the new influx of programmers integrating CC in their workflows.
You asked, we listened. The team has been cooking this for a while. Can't wait to hear what you think.
1
0
3
This is quite refreshing to see. Agents did make mistakes BUT they can be taught the right way to do critical domain specific things. Enter Agent skills. If you still are on the fence about skills being “just .md files with instructions lol”, you are missing out.
Since launching our AI for Science program, we’ve been working with scientists to understand how AI is accelerating progress. We spoke with 3 labs where Claude is reshaping research—and starting to point towards novel scientific insights and discoveries. https://t.co/WAvghBlbsC
0
0
0
Great insight. <<The flaws you see in your writing are invisible to everyone else.>>
I enjoyed answering questions from @c_a_dunlop about my approach to writing online. Here's my number one tip for people who want to publish more of their thoughts on the internet:
1
0
1
🤣🤣🤣🤣🤣🤣🤣🤣 relatable
0
0
2
> perf-hints : A language agnostic Claude Code plugin based on Jeff Dean & Sanjay Ghemawat's performance hints post. > /perf runs a 7-phase optimization workflow with back-of-envelope analysis. > /plugin marketplace add areu01or00/perf-hints >/plugin install perf > Link :
github.com
A language-agnostic engineering wisdom plugin for Claude Code based on Performance Hints by Jeff Dean, Sanjay Ghemawat's viral article - including performance optimization, code review, tes...
Performance Hints Over the years, my colleague Sanjay Ghemawat and I have done a fair bit of diving into performance tuning of various pieces of code. We wrote an internal Performance Hints document a couple of years ago as a way of identifying some general principles and we've
1
3
10
good for you, but isn't 1.5 million like half a researcher?
I'm happy to share that we (@AnthropicAI) are investing $1.5 million in support of the Python Software Foundation and open source security. Python powers so much of the AI industry. Supporting the folks that make our work possible is an honor.
11
5
436
Raw prompts : ### Dual Filesystem Instructions ``` The presence of this tool means that Claude has access to two computer filesystems: 1. The user's computer filesystem (this computer), which Claude can access using its Filesystem tools. 2. Claude's computer filesystem (the
1
0
2
Reverse engineered the macOS binary to see how Cowork functions. Below are the architecture details and Tooling prompts : The Core : Claude Cowork = GUI + Claude Code CLI + Linux VM Dual Filesystem is the Core. Claude has its own isolated computer (VM) , can access your
Introducing Cowork: Claude Code for the rest of your work. Cowork lets you complete non-technical tasks much like how developers use Claude Code.
1
3
9
Engram : A Hash Table for Language or L1 cache for language. Engram’s core idea is almost trivial in hindsight: bolt an O(1) lookup table onto the transformer. The pipeline is three steps: Slice: incoming token sequence is hashed into overlapping N-grams (e.g., 3-grams).
2
0
1
L1 cache for language is here
DeepSeek is back! "Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models" They introduce Engram, a module that adds an O(1) lookup-style memory based on modernized hashed N-gram embeddings Mechanistic analysis suggests Engram reduces the need
0
0
2
I hope your legal teams are ready
Introducing ChatGPT Health — a dedicated space for health conversations in ChatGPT. You can securely connect medical records and wellness apps so responses are grounded in your own health information. Designed to help you navigate medical care, not replace it. Join the
0
0
1
1. AI red teaming will be rampant. New attack vectors cultivating a billion dollar scam industry. 2. No amount of security and interpretability work will stop it. More outages. More leakages. More vulnerabilities. 3. Software quality degrades. Bar lowered. 3. General public
What are your predictions for AI in 2026?
0
0
1