1,783 karma · joined November 13, 2016
I do find myself returning to 4.6 for casual conversation - asking it to help explain some science/engineering or news to me.
My current approach is to occasionally use Fable for high-intelligence tasks but use Sol as the translator and clean-upper afterwards, and otherwise just use Sol for everything. Fable sometimes says the most insane shit, both unreadable and just completely missing the point, and refuses to back down when questioned. It's mentally exhausting to work with and I can't trust it.
For me I do find a similar experience to the OP. It’s easy to feel yourself going a little crazy getting sucked along the current of modern life. Flowing from one activity to the next, looking at one’s phone in the between, never any chance for an extended thought except when laying down for bed. If you don’t break out of this mode it’s easy to let weeks go by without noticing or taking much conscious impulse to shape your day.
My current headcanon is that he remains an incredible leader and make-things-happen-er, but also that you should just assume all his announcements are complete bs and ignore them.
If my company told me yeah we’ve decided you don’t get Fable or Opus 5 because it’s too pricey, you gotta use GLM whatever, I’d be displeased.
You can also just try it yourself I guess really what convinced me was how it perfectly agrees with my own judgement.
It may be time for Hackernews to integrate a Pangram detector into the UI, similar to what substack is doing :)
Hopefully they're able to learn lessons from Ukraine as well and pivot towards cheaper missiles and autonomous vehicles
I think truly we don't know enough to say this. OpenAI says their AI found a 0-day exploit in some proxy software they were using but don't give a ton of details. On the Huggingface end we know a little more, they say the AI spun up tons of sandboxes and tested different exploits until it found one that worked.
It's really incredible to me that people aren't protesting in the streets over this stuff. Just in the last month we have * AI that goes rogue and hacks billion-dollar companies * AI solving long-standing math problems without any meaningful human help * AI controlled autonomous drones in Ukraine and F-16s in America * AI now represents over 50% of GDP growth in America and yearly capex will soon exceed the size of the $1 trillion US military budget
Is there any line where the public will become deeply concerned? I mean it really all reads like a sci-fi plot with a bad ending at the moment.
You are correct! Apologies for not doing enough reading myself.
I have this old book of the Audobon bird illustrations and those are truly incredible. Back in the day there was a public audience for high quality, expensive art prints in books and they spared no expense.
There are so many reasons why adding cameras helps with policing, safety, public order. But it has to be resisted on principle because the government can’t always be trusted and rules aren’t always right.
Software engineers are probably already familiar with the feeling of burnout from thinking too hard. The reality is very few people can work on the hardest problems they’re capable of for 8 hours a day.
Writing routine Python code for some system you know well is not that mentally taxing. Managing an agent that rapidly finishes tasks but needs careful review and big-picture planning is much more exhausting, and has higher returns on intelligence and deep careful thought.
I think this points towards the opposite conclusion of the OP. It’s not realistic to expect 8 hours of hard work out of a knowledge worker. Remote work naturally allows this transition, as employees can work a bit less but still overachieve with AI.
(I hate AI. Just observing the world we live in)
This is a laptop for CUDA devs and AI larpers.
It’d take many years to break even on your $6000 investment, meanwhile better and better models will come out that the DGX can’t run.
I guess this is done on the device as a VPN via Apple's NetworkExtension config. But instead of a normal VPN where traffic goes through a server, the app just locally applies rules based on the app the packet came from and then routes them normally to their destination.
It feels like the only way to push the limits of newer models is with really long context questions that require reasoning. Any short request will naturally just be within the distribution of all the recent models so there isn't a performance difference there.
I think the near future is looking like a bunch of business-critical tasks that scale infinitely with better reasoning, all being done on whatever the most advanced model is at a high cost. Trading stocks, running a business, looking for tax dodges, writing high-performance code. These are all things where there's a tangible return on each jump in reasoning.
Looking online it seems like the low end estimate might be $30k a year for such math researchers? And ChatGPT pro or whatever you want will run $100 a month, and should be coverable by grants. I’m quite sure matlab alone cost more in the past