See, this is what I assumed when I mansplained the VR stuff to you last week.
I will mansplain metroidvanias instead.
See, this is what I assumed when I mansplained the VR stuff to you last week.
I will mansplain metroidvanias instead.
hell yeah
hahaha
that's the plot of my novel
Spent $199 on a CalDigit E5 and it doesn't even have ethernet.
My point is simply that many people legitimately find Gemini results to be better than anything they've gotten out of Google in >10 years.
There's a lot wrong with AI! But dying on the *particular* hill that LLMs can't possibly improve search at a reasonable compute cost does not seem wise.
Dragging Google for trying to build better search tools is also counterproductive. Web search was a computationally expensive workload when it was new. Hence all the ads!
Then consider the internet is 10-15x bigger than it was in 2016, in terms of total data. We really do need "smarter" search.
I believe those numbers amortize training costs. And models get used for good and bad things. Google/Gemini specifically has made efficiency and sustainability a focus.
There's a lot wrong with AI, but throwing tomatoes at people for what are ultimately cheap search queries is counterproductive.
Yeah, my work "upgraded" my CoPilot license because I needed the license to do some legitimate library science/archivist type stuff to un-disorganize vital SharePoint repositories and now in every Office app it is constantly asking to summarize and rewrite every single file and email I touch.
Yeah, all the "it makes you dumb/uncreative/etc" arguments still stand. But getting on Google's case for trying to un-break search with Gemini is one area where it's gotten a bit counterproductive. The internet is much larger than Google was built to index 10-15 years ago. We *need* smarter search.
Mos of what people on bluesky are doing when they complain about casual "LLM-as-search-replacement" use is like complaining about the water cost of almonds while eating a big juicy steak.
Ultimately, "AI" at large has a big impact, when you look at all the different types of AI workloads combined. But strictly providing text and web links, what Gemini Fast does, is computationally negligible. Especially when compared to people's streaming TV habits.
Data Centers Carbon Emissions at Crossroads: An Empirical Study
dl.acm.org/doi/abs/10.1...
also:
People are just conflating too many things when they talk about "AI", for reasons ranging from very legitimate to absurdly misinformed, while completely discounting the climate impact of all the other cloud/web services they use all day.
generating a 500 tokens, a page of text, is roughly equal to 7-10 seconds of video streaming, for any of the freely available text-only models as of a year ago. And those cheap models have only gotten cheaper since.
Simple search queries are just not the "AI" workload people should be arguing over.
Text and web-search AI queries are just not using the kind of compute that people on bluesky are claiming.
The real datacenter use driving up power consumption comes from image and video generation, along with "big data" business operations that have been growing at the same rate since before genAI
My point is that worrying about the climate impact of asking the bot for a pancake recipe and watching a ~20m video about pancakes instead is massively contradictory.
The footprint of Gemini for simple searches is just not worth thinking about vs everything else you do on the web all day.
the real climate impact from AI comes from image and video generation, or industrial-scale data operations, the latter of which isn't new to genAI
bluesky has deluded people into thinking that free 'fast' LLM queries use any more compute than anything else you do on the internet all day, including regular web search
the average person's daily netflix/youtube habit has 20-50x more climate impact than any amount of search-like LLM use
you clearly didn't read anything in my post
this novel should be a comic book
but i would have to make friends with someone who can draw
You could definitely nudge Gemini into feeling something like old Google with the right customization. A helpful link-finding bot, with negligible commentary. And Gemini Fast (aka not "Thinking" probably uses less compute at this point than a Google search did 5 years ago.
lololol
every day i am more convinced that ChatGPT is how the brawndo ended up on the crops
he's so dumb
levels of dumb rarely seen in public
but more accurately, the climate
so the pink one is from Godzilla x Kong: The New Empire i think, not from Monarch
definitely worth trying to get a free trial promo for Apple+ and plowing through a bunch of shows
oh right lol
that movie is so good