We’re still in the first inning. A model that is 50x more efficient means… more reasons to be bullish on AI True that there is all to play for though – and the incumbents — or those with biggest CapEx— can’t rest easy. Innovation wins, and it is arrogant to think that it only
RESEARCH
-
Perplexity adds US-hosted DeepSeek R1 reasoning
By
–
BREAKING 🚨: Perplexity released US-hosted DeepSeek R1 reasoning option 👀 https://t.co/Wm5KT4bmpO pic.twitter.com/yKnzCQCt7n
— 🚨 AI News | TestingCatalog (@testingcatalog) 27 janvier 2025BREAKING : Perplexity released US-hosted DeepSeek R1 reasoning option
-
DeepSeek’s Innovation Through Constraints and AI Development Alternatives
By
–
DeepSeek is the flip side of the same coin: It innovated because of, not in spite of, its constraints. And now that it has upended the assumptions of the dominant AI paradigm, we should reject its costly trade-offs and seek new ways to develop AI without so many harms. 24/
-
Scaling AI is Business Strategy, Not Scientific Necessity
By
–
As I said earlier, scaling was always more about business than science. Scientifically, there’s no law of physics that says AI advancements must come from scaling rather than approaches using the same or fewer resources. Scaling is just an incredibly easy-to-follow formula. 17/
-
DeepSeek challenges OpenAI’s narrative on AI development trade-offs
By
–
But DeepSeek shows – in the same moment as the Stargate announcement – that the trade off that OpenAI & co frame as wholly necessary is actually not. 15/
-

Synthetic Data’s Role in AI Training and Data Privacy
By
–
Ahead of #DataPrivacyDay, SAS' Brett Wujek explains the valuable link between the protection of personal data and synthetic data. Within just a few years, about half of the data used to train AI will be synthetic. This matters for you and your business. http://
2.sas.com/6012xB68w -
BAIR Alumni Develop New AI System Evaluation Test
By
–
Work led by BAIR alumni @hendrycks on a new evaluation test for A.I. systems. https://
nytimes.com/2025/01/23/tec
hnology/ai-test-humanitys-last-exam.html
… -
Text Compression Limits in Large Language Models
By
–
It’s hard to compress text deterministically beyond a certain point but LLMs seem quite far from that point. I’d certainly prefer such a model…
-
Correction: Model Performance Comparison O1 Versus O3
By
–
Petite coquille : plus performant que o1, pas o3. Ce qui ne change absolument rien au fond du désastre.
