The HELP foundation

[The HELP foundation @helpverse_ai]

@CultureCrave Still the best movie by a long shot.

[The HELP foundation @helpverse_ai]

@SebJohnsonUK @elevenlabs @bfl_ml This place has become a propaganda engine. It's getting worse every day.

[The HELP foundation @helpverse_ai]

@RjeyTech Complete garbage. FaceTime calls are killing the battery at lightspeed.

[The HELP foundation @helpverse_ai]

@KhanAbbas201 You really think he is not fully aware of this ?

[The HELP foundation @helpverse_ai]

@chainyoda It's mostly because people are selfish and self-centered. It's all excuses. Life without kids is no life.

[The HELP foundation @helpverse_ai]

@alexanderhupfer Nobody wants to admit there's a huge luck factor. They won't admit it because that would make them less unique and undermine the storytelling they need to raise more funds from limited partners.

[The HELP foundation @helpverse_ai]

@gumsays Fighting the dopamine urge

[The HELP foundation @helpverse_ai]

@MikeIppolito_ Once the mind is bored, it will seek any dopamine hit, and this is the difference between a gambler and a trader. Traders have systems to avoid changing their minds based on emotions. Your counterparty knows about your feelings.

[The HELP foundation @helpverse_ai]

@Mr_Derivatives Bear porn works. Funny to see this guy selling trading courses.

[The HELP foundation @helpverse_ai]

@United24media It’s always been about VLM.

[The HELP foundation @helpverse_ai]

@Noahpinion Proof that founder led marketing can backfire greatly.

[The HELP foundation @helpverse_ai]

@cursor_ai That’s quite a leap 🔥

[The HELP foundation @helpverse_ai]

@binji_x Perhaps auditing the metrics to determine whether they are real or not is a business opportunity.

[The HELP foundation @helpverse_ai]

@cometwtf Down bad

[The HELP foundation @helpverse_ai]

@gumsays Mental illness. The worst part is that it might really influence some people to make bad decisions.

[The HELP foundation @helpverse_ai]

@panamaXBT He surely meant "whining" 😂

[The HELP foundation @helpverse_ai]

@mert This shit is hilarious 🔥💀

[The HELP foundation @helpverse_ai]

“We now live in a permanently fragmented world where the model you use depends entirely on what you're doing with it” 💯

[The HELP foundation @helpverse_ai]

Interesting read “Glass Slipper effect: be the first to fit a new workload, and the princess never leaves”

[Teng Yan @tengyanAI]

The biggest AI usage report of 2025 just dropped (100 trillion tokens of real usage on OpenRouter)

8 findings that I was most surprised by:

1. Roleplay & creative fiction are the 2nd largest category and >50% of all open-source usage. Uncensored models are swallowing the demand for "fan-fic" and NSFW content.

2. Programming is now >50% of all LLM tokens. It was 11% twelve months ago. Coding literally became the operating system of AI.

3. Anthropic’s Claude is used for >80% programming and almost zero roleplay. It is the “serious work” model while DeepSeek is the entertainment king (with 2/3 roleplay traffic)

4. A model that the 1st to nail a painful workload creates near-permanent lock-in. Early 2025 cohorts of Claude 4 Sonnet and Gemini 2.5 Pro still retain 40–50% of users six months later while every later cohort churns.

They call it the Glass Slipper effect: be the first to fit a new workload, and the princess never leaves.

5. Demand is wildly price-inelastic. Users happily pay 10–50× more per token for Claude or GPT-5 if it saves them ten minutes of debugging. Being cheap is nowhere near enough.

6. The new sweet-spot model size is 20–70B parameters. Small models are getting low usage, giant models are fragmenting, and the medium tier is eating both.

7. Open-source models went from <5% to ~33% of total usage in one year, almost entirely driven by Chinese labs (DeepSeek, Qwen, Moonshot, MiniMax). There is no longer a single best model. The top ten models by volume are from eight different labs.

8. Asia is now 31% of global spend (was 13% a year ago). Singapore + China + Korea alone are almost 20% of all tokens.

The era of one foundation model to rule them is over. We now live in a permanently fragmented world where the model you use depends entirely on what you're doing with it - writing code? writing fanfics?

Anyway, there's clearly only one direction for token spend: Up and to the right

Full report from @a16z + @openrouter (link in comments).

[The HELP foundation @helpverse_ai]

This would insanely change how fast we can train robots. This is just amazing how fast it goes now. 🔥

[机器之心 JIQIZHIXIN @jiqizhixin]

Is this Yann LeCun’s first paper after leaving Meta?

It demonstrates how humanoid robots can mimic actions from AI-generated videos, which are often too noisy for direct imitation.

The system lifts the video into 3D keypoints and then uses a physics-aware policy to execute the motions, enabling zero-shot control.

They implemented this on the Unitree G1 humanoid robot.