I do not represent the views of any company. Floats | Weights | Based | C++ | C | IO | Latency | SW @ HW Middleman. Reversing Ex SSE at MSFT
India
Joined April 2022
- Tweets1.9K
- Following925
- Followers57
- Likes1.6K
TPU and flash models are a disgrace to humanity
Today, we're bringing OpenAI models to Google Cloud's model garden.
We're now the only hyperscale cloud provider offering access to all three leading closed source LLMs—alongside 200+ other models—on a unified enterprise platform → goo.gle/4zJ0VZQ
I have launched check_submission.sh in the background to verify the codebase after adding the requested comment.
stuff gemini says -- this is what ceos in build mode or vibe coders ship
your models suck because it is a feature vector text predictor and in last 18 years of FREE RAM - low or -negative quality videos or blogs have ruined the web. What is the need for another blog on algorithms when one book is there?
It is only a matter of time before companies stop paying content creators based on human-driven "likes" and begin valuing the intellectual quality of their videos instead. While quantity has its place in scaling, it simply doesn't apply here.
Gemini 2.5 started as an exceptional model. It isn’t enough for accurate data to exist; we must also eliminate the bad. The influx of diluted technical content and clickbait-driven influencer videos on YouTube has significantly compromised the quality of the model. Futuure muse
explicit SequenceData(std::vector<int> prompt_token_ids)
: token_ids(std::move(prompt_token_ids)), prompt_len(static_cast<int>(token_ids.size())) {}
explicit SequenceData(const py::array_t<int32_t, py::array::c_style> &prompt_token_ids) {
Slop c++
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/DeepI…
1/4
Another middleman
A developer in India should not need a dollar card and twenty accounts to use AI.
We are building the fix. It is called Rapid Router.
Pre-launch is live: rapidrouter.ai
Knock knock, who? Knock knock, who? Who? Knock knock. AI still cannot do multi-threading.
Soon we will have ChatGPT music and drama and long 100-minute videos where you will ask a -`... and ChatGPT will take its sweet time to explain the answer, not that text and codex useless stuff.
The best ai invention. Not that codex shit it's only a matter of time before countries and companies will be begging for ChatGPT voice.
The best ai invention. Not that codex shit
Say what you want, The Chinese CEOs seldom talk about ROCE, ROE, etc.. Look at Indian CEOs, all they talk about is engineering challenges, hard problems, impossible problems with solutions /s
People were already writing slop Pre AI in GC assisted languages with IDE. Yes, RAM was free in cloud era and too many key value pairs and paxos lovers made money, AGI will delete them all
given enough gpu open ai can solve math problems -- but it still cannot and will never get multi threading right and will never get it right ever it still gets cannot
People who claim that AI hallucinations are a thing of the past clearly have never spent 10+ hours struggling with a complex problem from scratch. Hallucinations are fundamentally different from model laziness—and the experience speaks for itself.