Weekly Top Picks #95
DeepSeek-R1 = o1 / OpenAI and FrontierMath / Writers coming to grips with AI / Sam Altman's hype-dehype tactic / Video deepfakes / Chollet x (DL + GOFAI)
The week in AI at a glance
DeepSeek-R1 on par with OpenAI o1 (and open source): Smaller, cheaper, but just as good. DeepSeek-R1 confirms that open source has caught up to OpenAI.
OpenAI had exclusive access to FrontierMath: OpenAI’s collaboration with and funding of EpochAI raises questions about fairness and transparency.
When professional writers come to grips with AI’s skill: Paul Schrader is awed by AI’s writing ability, a shift writers resist but must eventually accept.
Sam Altman’s posting style in two tweets: Altman’s hype-dehype tweets reflect his effort to balance public perception and maintain momentum for OpenAI.
Video deepfakes: state of the art: Hunyuan’s video-to-video model can make good deepfakes, which, contrary to popular fears, are not really about deception.
F. Chollet to combine deep learning and symbolic AI: Ndea aims to merge deep learning with symbolic reasoning for AI capable of independent discovery.
The week in The Algorithmic Bridge
(PAID) Weekly Top Picks #94: OpenAI o1 isn't a chatbot / Invisible progress / Prepare for what's coming / OpenAI agents / Is AI slop good? / Against false positives / The risk of being anti-AI / The denial of AI / AI music
(FREE) How AI Is Quietly Changing What It Means to Be Alone: Drawing from Derek Thompson’s insights on rising solitude, I explore how AI might intensify this trend, making it the norm with personalized entertainment and AI companions. Rather than fostering connection, AI risks fragmenting shared culture, replacing communal experiences with tailored content. This isn’t a call to reject AI but a warning about the cost of further isolation: trading the shared messiness that defines our humanity for solitary perfection.
(FREE) This Rumor About GPT-5 Changes Everything: What if GPT-5 is real, but OpenAI kept it internal for greater ROI—not in money, but in something else? While there are no leaks or insider info, public evidence suggests GPT-5 is already shaping smaller, cheaper models (GPT-4o? o1? o3?). Like Anthropic’s decision to distill Opus 3.5 into the cost-effective Sonnet 3.6, OpenAI may use GPT-5 as a data mine to refine models without public release. As OpenAI chases AGI, they might prioritize internal breakthroughs over public access.