AI: Weekly Summary. RTZ #632

Feb 15, 2025

OpenAI’s simplified product roadmap: Founder/CEO Sam Altman announced new details and simplification of OpenAI’s product roadmap that could end up being a trendsetter for peers and competitors. Specifically, OpenAI’s next LLM AI model is GPT-4.5, which was code named ‘Orion’ last fall internally. It’s potentially out in the next few weeks. That will be the last model before the company’s ‘chain of thought’ reasoning technologies like Deep Research, Operator et al, are all built into the company’s next model later this year, dubbed GPT-5. Particular focus is on not having users needing to choose from almost half a dozen models and sizes from a dropdown menu, every time they want to use ChatGPT. Also simplified is the company’s pricing at various tiers and query quotas for its various products. These steps should help the company transition from developers and early adopters, to more mainstream consumer and business users later this year. More here.

Google YouTube’s AI Roadmap for 2025: YouTube CEO Neal Mohan rolled out the company’s 2025 roadmap, with a key milestone highlighted: that YouTube on TV has surpassed mobile as the ‘primary device’ for mainstream viewing. This has big implications for the media business worldwide, as the multi-hundred billion dollar ‘Creator’ Economy, fueled in particular by YouTube’s leadership, fuses with mainstream media online. TikTok by Bytedance, and ‘Reels’ by Meta Instagram/Facebook of course are another manifestation of this trend. The trend also has big implications for other stakeholders like advertisers and global content publishers. AI technologies are already accelerating the content production on video services worldwide, with YouTube in pole position. More here.

Making AI look like it’s ‘Thinking’: As AI companies move from ‘chatbots’ to AI Search, Reasoning and Agentic capabilities, a ‘hot’ technique is to make it look like the AI ‘is thinking’. The technical imperatives of doing this mean innovations throughout the AI tech stack, from the back end training/inference compute, to the front-end UI/UX. Companies including OpenAI, DeepSeek, Perplexity and others are leaning in here, with DeepSeek’s reasoning R1 UI/UX being used as a model in particular. These techniques also extend to voice driven multimodal AI capabilities, being rolled out by Google, OpenAI, Amazon, Apple and others. Anthropomorphizing AI is now an active ‘feature’, not a ‘bug’. More here.

Apple partners with Alibaba for AI in China: Apple is overhauling its iPhones in China, with AI technologies from Alibaba and Baidu. The systems will be compliant with regulatory rules on censorship and content blocks from the government. CEO Tim Cook had cited a lack of its Apple Intelligence AI features in China as a reason for softer smartphone sales in that market. The strategy is a customized version of its AI strategy in the US, with its partnership with OpenAI’s ChatGPT and potentially other AI systems. Apple will also store China user content in the country, leveraging a partnership with state-affiliated GCBD to operate its iCloud online storage service. Apple apparently also tested DeepSeek’s AI technologies, but the smaller company might not have the scale to support Apple’s China operations. More here.

Early days for AI wave in 2025: OpenAI’s Sam Altman reaction to the January global web traffic data this week was telling. Commenting on no. 6 ranked OpenAI’s 2.66% share vs almost 50% for Google’s core Search and YouTube properties (no. 1 & 2), and Meta’s 10% at nos. 3, 4, & 5, founder/CEO Sam Altman pithily said on X/Twitter: “man, still a long way to go to run down google”. Even with market leading LLM AI and Reasoning products with over 300 million weekly users, and 15 million ChatGPT paying subscribers. Highlights the early days to AI mainstream usage. It’s so easy to think we’re so far along the AI Tech Wave two years after OpenAI’s ‘ChatGPT moment’ and hundreds of billions in AI Infrastructure Data Center and Power capex to date. More here.

AI: Reset to Zero