Artificial Intelligence thread

tphuang

tphuang

General
Staff member
Super Moderator
VIP Professional
Registered Member
https://twitter.com/i/web/status/2017294075620184442

Please, Log in or Register to view URLs content!

so not only DeepSeek, but ByteDance and Alibaba are launching new models around CNY.

and they are all rushing out red envelopes to get people to use during new year period.


I've been using Kimi a lot in the past couple of days. Really good. I actually find it to be the best free access LLM I tried online. Much better than current DeepSeek and Gemini-3 (at least the free version available online).

This is literally the first time you had a Chinese AI model available that's almost as good as the latest American moderns.

I remember when R1 came out, o3 came out right afterward and o3 was just leagues ahead of R1 in the answers. As far as I can see, OpenAI isn't holding things back anymore so GPT 5.2 is the best they got and it's only a little better than Kimi based on the scores.
 
Wrought

Wrought

Captain
Registered Member
Long article about successful efforts to cultivate the STEM education pipeline. Written by a dropout.

The classes quickly became a standard feature for thousands of schools — and the results were impressive. As the years passed, Chinese teams started to sweep most of the gold medals at Olympiads, far exceeding their rivals. In 2025, the Chinese national teams sent a total of 23 contestants to the Olympiads: 22 came home with gold medals. Starting in the 2000s, university admissions were reformed, giving more flexibility to colleges to allocate places without relying solely on the results of the gaokao. National competitions were set up for students at the end of their sophomore year of high school. Those who won top prizes in the national exam could receive direct admission to one of the 985 Project universities, China’s 39-member Ivy League equivalent.

The chance to skip the gaokao was a strong incentive for students to participate in the genius stream. The traditional pathway for high-school students in China is three years of study in the gaokao’s mandatory subjects of Chinese, English and Maths, as well as three more chosen subjects from physics, chemistry, biology, history, geography and politics. Exams in all six subjects are taken at the end of the third year. Genius-class students, on the other hand, focus on their “competition subjects”. A student competing in the International Physics Olympiad, for example, needs to not only finish three years of high-school physics but also at least half of the college-level syllabus, in order to be competitive enough to take the national exam. The very dedicated might not study much else at all.
Click to expand...

Please, Log in or Register to view URLs content!
 
T

tokenanalyst

Lieutenant General
Registered Member
Step 3.5 Flash is our most capable open-source foundation model, engineered to deliver frontier reasoning and agentic capabilities with exceptional efficiency. Built on a sparse Mixture of Experts (MoE) architecture, it selectively activates only 11B of its 196B parameters per token. This "intelligence density" allows it to rival the reasoning depth of top-tier proprietary models, while maintaining the agility required for real-time interaction.

  • Deep Reasoning at Speed: While chatbots are built for reading, agents must reason fast. Powered by 3-way Multi-Token Prediction (MTP-3), Step 3.5 Flash achieves a generation throughput of 100–300 tok/s in typical usage (peaking at 350 tok/s for single-stream coding tasks). This allows for complex, multi-step reasoning chains with immediate responsiveness.
  • A Robust Engine for Coding & Agents: Step 3.5 Flash is purpose-built for agentic tasks, integrating a scalable RL framework that drives consistent self-improvement. It achieves 74.4% on SWE-bench Verified and 51.0% on Terminal-Bench 2.0, proving its ability to handle sophisticated, long-horizon tasks with unwavering stability.
  • Efficient Long Context: The model supports a cost-efficient 256K context window by employing a 3:1 Sliding Window Attention (SWA) ratio—integrating three SWA layers for every one full-attention layer. This hybrid approach ensures consistent performance across massive datasets or long codebases while significantly reducing the computational overhead typical of standard long-context models.
  • Accessible Local Deployment: Optimized for accessibility, Step 3.5 Flash brings elite-level intelligence to local environments. It runs securely on high-end consumer hardware (e.g., Mac Studio M4 Max, NVIDIA DGX Spark), ensuring data privacy without sacrificing performance.
1770047569263.png

Please, Log in or Register to view URLs content!
 
You must log in or register to reply here.
Top