hi bud buds,
last week, openai launched gpt-5 to all chatgpt users.
many are calling it disappointing, saying changes feel less transformative than prior launches. for example, many highlighted improvements are single-digit percentage gains (e.g., “we’ve reduced rates of deception from 4.8% for o3 to 2.1% of gpt-5”).
personally, it feels like a good launch, just not revolutionary, which we’ve come to expect from openai. however, revolutions can’t happen every few months.
i spent a couple hours digging through gpt-5 launch announcements / tweet replies; here are the key takeaways for pms:
⭐ enjoy cheaper prices: 5–10x cheaper than other flagship models, with slightly better performance 2
⭐ watch for fewer hallucinations: ~45% less likely to contain factual errors than gpt-4o ¹
upload more screenshots / pdfs: handles images, charts, presentations, and diagrams more accurately ¹
do more ai prototyping: altman says gpt-5 excels at spinning up entire software apps on demand 3
personalize your chatgpt: choose personalities for your chatgpt (cynic, robot, listener, nerd) to suit your needs ¹
more professional tone: less overly agreeable, fewer unnecessary emojis; feels more like chatting with a phd-level friend than “with ai” ¹
enjoy better agent performance: i still don’t know any pms using agents regularly, but gpt-5 demonstrates improved instruction following + tool use for multi-step, multi-tool (aka: agentic) requests ¹
tweak settings less: queries dynamically routed to right models and tools based on query type, complexity, and goals ¹
as you can see, nothing quite “game-changing” for pms. the biggest updates are pricing (cheaper) and hallucinations (fewer), but neither feels transformative. the strongest bull case i’ve seen for gpt-5 is “fewer hallucinations are a huge unlock for enterprise adoption,” which might be true, but still feels modest by openai’s usual standards.
perhaps the clearest sign that openai recognizes gpt-5 is technically underwhelming: they’re moving the goalposts and shifting the spotlight to non-technical “wins” like pricing and “taste.”
making gpt-5 affordable is indeed noble and a huge win for society. however, it is entirely controllable by openai and independent of technical success. similarly, while “taste” is great, it is subjective and can’t be refuted. including in a press release that “the vibes are really good” just doesn’t feel up to the bar openai has set in the past.
what does this mean for openai and their agi mission? to me, it suggests llms are improving at a rapidly decelerating pace. i’ve always been skeptical of sama’s claims that “agi” is imminent, and this feels like further validation that we’re still a long way off.
have a great week, y’all. i love you.
- dj
last week’s biggest product releases
major
openai also released gpt-oss-120b and gpt-oss-20b, its first open-weight models in 5 years. oss-20b is small enough to run locally on end devices, opening massive potential for developing markets and privacy-sensitive industries (e.g., healthcare, finance)
x is adding ads to grok’s responses, letting advertisers pay to appear in chatbot answers
xai debuted grok imagine, an ai image/video generator that allows nfsw content
character ai is launching a social feed, letting users share chat snippets with ai characters, and even have their characters debate topics on a livestream
hell yeah
x is restoring vine’s archive, making old videos accessible again, and i couldn’t be more stoked. here’s an all-time favorite to celebrate
locket rolled out celebrity lockets, letting celebs send updates directly to fans’ home screens. absolutely love the idea, but hard to see patreon not copying this quickly
meta added instagram maps, similar to snap map for social location sharing
lyft plans to launch autonomous vehicles in germany and the uk in 2026
google is revamping google finance, adding ai research, live news, and advanced charting tools
solid
google added guided learning, a step-by-step ai tutor in gemini
google is making notebooklm available to users 13+ (previously 18+) with ai study features like converting notes to podcasts / videos
spotify expanding audiobooks+ access to family / duo plans
whatsapp added anti-scam features, including scam detection in chats and alerts for suspicious contacts
roku launched howdy, a $3/mo ad-free video streaming service
other llm news
anthropic dropped claude opus 4.1, improving agentic tasks, coding, and reasoning at the same cost
google deepmind launched genie 3, a “world model” for training ai agents in 3d environments (til: a world model = can “perceive” the real, living world)
google launched jules, an agent-based coding tool powered by gemini 2.5 pro
idk tbh
the browser company launched dia pro, a $20/mo ai browser subscription with unlimited chat + skills. ai browsers are exploding in popularity, but i haven't loved dia yet tbqh
that’s it for this week; thanks for reading. if you enjoyed this issue, please consider sharing it with a friend so i can get rich and buy my wife a nice designer gown.
love, dj