Our 2025 Tech Predictions and Resolutions + We Answer Your Questions

Summary

• Benchmark parity did not neutralize ChatGPT’s consumer lead: Gemini topped Chatbot Arena, but ChatGPT reported 300 million weekly users and Casey said Google had disclosed nothing comparable. Casey Newton’s 2024 thesis that “quality differences will matter less and less, and distribution will matter more” was only half-right—Gemini caught up on some benchmarks, yet ChatGPT became the cultural default despite lacking Google’s distribution channel.
• Casey’s highest-confidence 2025 call is an AI culture war, resolved by a congressional hearing over a chatbot’s response. The tinder includes perceived political bias, children and adults forming intimate chatbot relationships, and the possibility of a right-leaning model built from Llama; he does not know the spark, but says “everything is in place for this to have a moment in 2025.”
• Kevin Roose expects a newly launched meme coin to touch $100 billion before crashing as a crypto-friendly Trump administration accelerates speculative gambling. His template is Hawk, which briefly approached a $500 million market cap before losing more than 95% within hours despite having no underlying value. Casey agrees that “all bets are off,” while stressing that the gambling expansion is harmful and destructive.
• Waymo’s geographic expansion could turn autonomous driving from a Bay Area curiosity into a mainstream cultural phenomenon. Service already covered San Francisco, Phoenix and Los Angeles, with Atlanta, Austin and Miami coming; Casey’s resolution criterion is an SNL sketch, while Kevin thinks New York deployment is the real catalyst because so much media remains concentrated there.
• Kevin gives medium confidence to Apple acquiring Snap, pairing a struggling social company with Apple’s smart-glasses ambitions. Snap’s stock was down about 23% year to date, it had conducted layoffs, and a friendlier M&A climate could help—but Apple would inherit difficult privacy, security, child-safety and CSAM responsibilities that it currently leaves to App Store developers.
• Casey’s low-confidence corporate call is that X will be folded into xAI because X’s long-run value is its training data. xAI was already valued at $50 billion—above Twitter’s $44 billion valuation when Musk acquired it—and Elon Musk planned to expand its supercomputer toward one million GPUs. The caveat is that Musk already shifts resources between companies, so a formal merger might change little initially.
• Kevin thinks OpenAI might declare AGI in 2025 partly to escape its Microsoft obligations, not merely to claim a technical milestone. Once AGI is reached as defined by OpenAI’s nonprofit board, Microsoft could lose access to future models; Casey counters that negotiations over removing that clause imply OpenAI still needs Azure and Microsoft’s infrastructure. Kevin’s own hedge: the partnership simply feels “very messy and like it may explode at some point.”
• The episode’s practical AI thesis is strongest where errors are tolerable: novice coaching, idea iteration and early-stage learning rather than mission-critical judgment. Casey used Claude as “a journal that talks back to you” while learning meditation; Kevin found AI useful from zero to beginner-level poker but weak at advanced strategy. Conversely, image proportions remain unreliable, training-data provenance is opaque, and voice authentication should be abandoned because short audio samples can enable convincing clones.

Deep dive

Not yet available upstream; scheduled sync will retry.