If @clem comments on this post within the week with a task for me to classify with text and a paramater size for the model, within 48 hours I will create a new dataset, train the model, and post it all. Paramater size must be bellow 50 million params, and task can be text only, and genuinely possible.
if I complete it, Clem must tell everyone on Twitter/x to follow me on huggingface and link it
why did 36 people unfollow me 😭 we are back in the hundreds. if you become my 500th follower and have proof I'll give you 5 dollars worth of openrouter credits as an API key
• **Qwen3 update** – Alibaba’s Qwen team released an update to its Qwen 3 model. The latest Qwen3‑235B‑A22B‑Thinking‑2507 has 235 billion parameters with 22 billion active (MoE), supports a 256 k context and introduces a reasoning mode【887277669479733†L94-L103】, making it “agent‑ready.”
• **Qwen3‑Coder** – The 480 billion‑parameter Qwen3‑Coder activates 35 billion parameters and can handle 256k‑1M token contexts. It tops SWE‑Bench and CodeForces benchmarks and can generate, refactor and debug code across languages【813759831752062†L31-L69】【813759831752062†L96-L113】.
• **HRM** – MIT’s Hierarchical Reasoning Model uses separate high‑level and low‑level modules; with only ~27 million parameters and 1 k training examples it outperforms chain‑of‑thought LLMs on reasoning tasks【961622371027367†L74-L93】.
• **ASI‑Arch** – An arXiv paper presents ASI‑Arch, an autonomous AI research system that ran 1 1773 experiments (~20k GPU hours) and discovered 106 new linear‑attention architectures【159627632585686†L49-L73】.
Other headlines: OpenAI & SoftBank are building a compact data center【594568962713797†L92-L99】; Lloyds Bank launched the Athena AI assistant【594568962713797†L218-L226】; Yahoo Japan plans daily use of generative AI【594568962713797†L73-L79】; and AI‑agent tokens like FET, Virtuals and OriginTrail are powering Web3 automation【178973618089356†L24-L63】.
### ChatGPT Agent Corner OpenAI’s ChatGPT Agent unifies browsing, coding, document analysis, slide creation and scheduling into one interface【619815554729605†L14-L24】. It scores 41.6 % on Humanity’s Last Exam and 45.5 % on SpreadsheetBench【619815554729605†L42-L52】, and is available to Pro, Plus and Team users【619815554729605†L65-L69】.
We zoomed pass 70, 80, and 90 followers in a day. That is insane.
As a thank you, ive published one of my largest and most expensive dataset (rip my 2 dollars, most of my datasets are free to make) to the hub. To be used for text summarization in english, check it out! ProCreations/simple-summaries
New dataset I made! ProCreations/quality-fiction This is a synthetically made dataset with about 200 examples. It contains high quality fiction/fantasy story to use for AI training.
What is AI? Some weird wizards at big companies summoning brains? Humans behind the scenes just responding to requests? Nope. It’s literally JUST math. Some people know this already, but not what actually goes on behind the scenes.
Uggghhh what a busy week. Been up all night fixing Intellite and I got some progress I wanted to share.
ALL errors we experienced (there’s a lot): Gradient vanishing (AI dies) Gradient explosion (AI barely learns and destabilizes) Activation Explosion (AI dies) Layer Scale Crushing Signal (AI dies) Activation Explosion (again) (AI dies) All fixed, but now it keeps overfitting on datasets due to a dataset handler error 😩
