cmeinel

The gap between Apple and NVIDIA today is $407B. This compares with @geoffodlum reporting $239B March 16 here https://www.randforecastinginitiative.org/comments/161803

There are new competitors who do great on the GenAI benchmarks without using the top of the line Nvidia chips, indeed not many chips of any sort.

https://www.analyticsvidhya.com/blog/2025/01/deepseek-r1-vs-openai-o1/

https://www.reuters.com/technology/us-looking-into-whether-deepseek-used-restricted-ai-chips-source-says-2025-01-31/

Still no sign of preparations for a test at https://www.38north.org/

Sometimes, less is more. Is that time arriving in the world of users of NVIDIA chips?

Dateline 2-28-2025 S1 is based on a small, off-the-shelf AI model from Alibaba-owned Chinese AI lab Qwen, which is available to download for free. To train s1, the researchers created a dataset of just 1,000 carefully curated questions, paired with answers to those questions, as well as the “thinking” process behind each answer from Google’s Gemini 2.0 Flash Thinking Experimental.

After training s1, which took less than 30 minutes using 16 Nvidia H100 GPUs, s1 achieved strong performance on certain AI benchmarks, according to the researchers. Niklas Muennighoff, a Stanford researcher who worked on the project, told TechCrunch he could rent the necessary compute today for about $20.

The researchers used a nifty trick to get s1 to double-check its work and extend its “thinking” time: They told it to wait. Adding the word “wait” during s1’s reasoning helped the model arrive at slightly more accurate answers, per the paper.

Source: https://techcrunch.com/2025/02/05/researchers-created-an-open-rival-to-openais-o1-reasoning-model-for-under-50/

Recent rollout of Blackwell could slow chip deliveries and introduces more uncertainty.

Dateline 2-17-2025 Nvidia CEO Jensen Huang said on the company’s last earnings call in November that Blackwell production “is in full steam.” But many analysts don’t expect those Blackwell shipments to hit high volume until the second half of the new fiscal year, which could mean a disappointing forecast from Nvidia for its April-ending quarter.

Source: https://www.wsj.com/finance/stocks/deepseek-did-nvidia-a-favor-508c7e27

China is far too vulnerable to economic retaliation. Given its currently shaky economy, and by comparison with how easily mass protests developed against its Covid restrictions, I doubt that its leaders would dare to even just blockade Taiwan within six months.

Dateline 1-27-2025 China’s economy has slowed and faces a multitude of challenges: a real estate crash, banks burdened with bad loans, youth unemployment and underemployment, highly indebted local governments, and a population both aging and declining. On top of that, foreign investment is dropping. Indeed, not just falling but reversing.

Source: https://www.cato.org/commentary/chinas-faltering-economy-investment-beijings-challenge-home#

See "Rapid  in silico directed evolution by a protein language model with EVOLVEpro,"  https://www.science.org/doi/10.1126/science.adr6006  Dated

Editor’s summary
Altering protein function or stability through mutations is a powerful tool for engineering, but it is experimentally demanding to explore even a small fraction of potential mutations. Jiang et al. developed a computational approach called EVOLVEpro that uses recently developed protein language models within the framework of a top-layer regression model. A small number of experimental observations can guide protein evolution toward the desired functional outcomes and avoid stable, but nonfunctional, deadends. The authors demonstrate the capabilities of the method for evolving desired functions in six different example cases relevant to current molecular biology challenges, including evolving a miniature CRISPR nuclease, optimized antibodies, and an improved T7 RNA polymerase. —Michael A. Funk   

Still no sign of preparations for a nuclear test. Source: https://www.38north.org