whatever Google is doing with compute, what we know is that they nerfed even the flash models in chat. you can talk about something, comparing items A and B, then next message ask for another attribute for comparison, and it would ask "what are you comparing?". Going back several messages seem to literally reset the context? it would forget everything that had been discussed thus far.

hi @Google whats the reason why you do not surface the map & location information when its in AI search mode? e.g. if i look up a restaurant, in classic search, this would bring up the maps modal to show me where it is and information such as operating hours, etc, saving it for future "want to go". but in the AI search results, this stop showing up. is this intentional design choice? if so, was this to increase friction and thereby increase user use time? because this is one of the worst ux ive ever experienced, a significant downgrade of ux compared to classic search.

strangely, i actually think muse spark is benchmaxxed, if you use muse code you would find that its actually not that well rounded and performance is definitely not same tier as Sol. gemini 3.8 flash is more preferable for me.

the fear of it being benchmaxxed feels overblown at least for my use cases.

thanks @JeffDean for everything! So much of what I learned in school were what you worked on. You inspired me to always think bigger and challenge the working paradigm, and always look for ways to do something better! Look forward to you & rest of the discovery loop team’s next big thing sir!

youtube.com/watch?v=8hfpLa5w… - no comments the personnel changes being positive / negative, but 3 / 4 folks from this video recorded just 2 months ago are no longer at Google

idk, obviously everyone is selling and from latest round of earnings call it sounds like it’s about how you market the capex, but kinda seems wild that reaction to google and meta are negative? meta especially if zuck is claiming some shit like 1p roic will be > selling compute?

no idea why people are trashing google when the information coming out all seem to point to strong performance from them? model sucks due to 1. their focus on serving across their surfaces well ( speed and cost and ux ). 2. not enough capacity it seems? not enough capacity is due to a boom in gcp, which is positive. frontier (coding) model development will remain challenging for folks not getting good user data, but distillation and efficiency optimizations are core to google DNA, and recent models - tho not just simply distilled, is still largely relying on distillation, so feel like they should never be that far behind as long as they are not compute constrained?

nothing against josh, but gemini product team is honestly so behind if they are asking and getting these kind of answers. it’s like they don’t use the app at all and have no idea what they have on their own property is broken. let alone advanced features like going back in chat or integration with other products through mcp / agentic capabilities. or personalization doesn’t work or why do we have notebooks / projects when they aren’t able to reference past chats properly. the model is bad, the product is bad. on the other hand i have heard multiple folks telling me gemini and its related teams are giving sky high offers for just swe positions.

$meta is down because the trailer for the new movie dropped.