This is an interesting article that sums up some of what we already know—but summarizes it again: OpenAIs cost-cutting with GPT-5. When you consider that ChatGPT is used by 700 million weekly active users, of whom only 3% are paid users, it reveals the enormous scale and, at the
MARKET TRENDS
-
AI Model Development Increasingly Difficult for Highly Capitalized Companies
By
–
Some signs that catching up in the AI model space is rapidly becoming challenging for even the most highly capitalized companies.
-
Deep Think Rate Limits Doubled Through Competition
By
–
Deep Think rate limits have been doubled. Competition brings us great advantages.
-
AI Data Centers: Massive Energy Consumption
By
–
Meta and OpenAI are building data centers equivalent to 5 nuclear power plants.
AI devours electricity and office value. This is the new "headquarters" of AI giants. My new article co-written with @dr_l_alexandre in Trends. -
Top 10 AI Video Generation Models Ranked by User
By
–
J'ai eu occasion de tester tous les modèles de génération vidéo par IA, voici mon classement top 10, du meilleur au pire :
— VISION IA (@vision_ia) 14 août 2025
1. Google Veo 3
2. Midjourney video HD
3. Wan 2.2
4. Kling AI
5. Aleph de Runway
6. Hailou
7. Seedance
8. Framepack
9. Sora
10. Adobe firefly video… pic.twitter.com/3kurfkliw7J'ai eu occasion de tester tous les modèles de génération vidéo par IA, voici mon classement top 10, du meilleur au pire : 1. Google Veo 3
2. Midjourney video HD
3. Wan 2.2
4. Kling AI
5. Aleph de Runway
6. Hailou 7. Seedance
8. Framepack
9. Sora
10. Adobe firefly video -
Pro AI Models Excel at Hard Problems Requiring Expert Evaluation
By
–
The pro models (GPT-5 Pro, Gemini 2.5 Deep Think, Grok 4 Heavy) can be impressive in ways that are hard to see. They take a lot of time to answer questions & are built for very hard problems that require expert evaluation. That is a narrow, but, also very valuable, problem space.
-
DeepSeek r2 Delay: Ascend Chips Underperform Against Nvidia H20
By
–
The reasons why DeepSeek r2 has been delayed: Despite a team of Huawei engineers working on site, it was not possible to carry out a functioning training process with Ascend chips. Nvidia H20 chips are still significantly superior to Chinese alternatives in terms of
-
AI Investment Risk: 25% CAGR Growth with Binary Outcomes
By
–
AI is growing at 25%+ CAGR
Generational wealth is being built in AI. If you invest in AI firms in 2025: you will either build massive wealth or you will wipe out 90% of your capital. Many are comparing the current AI revolution (2025) to DotCom bubble (early 2000s) – VCs -
Google Rolls Out Memory Feature in Gemini AI
By
–
AI NEWS: Google is finally rolling out memory in Gemini Plus, more news from xAI, Anthropic, Tencent, Higgsfield AI, and Liquid. Here's everything you need to know:
-
DeepSeek R2 Delayed Due to Huawei Chip Restrictions
By
–
DeepSeek R2 delayed because they are forced to use Huawei chips