@aliceisplaying "anthropic quantized my model" is the new "my phone records me talking and gives me ads about it"
English
nake
55 posts
ppl are failing barely-reading-between-the-lines 101
everyone knows Anthropic is in an insane compute crunch and no employee can ever tweet this outright *which is normal*
but one thing they have said so many times very straight: they serve the same model at all times
English
when I first moved to SF, I was shocked by how expensive everything was. $18 for a burrito with guac? $80/month for a gym?
Then found some affordable places like Saigon Sandwich $6 for sandwich, $4 Vietnamese coffee.
I started keeping a list of actually affordable spots:cheap eats, budget gyms, grocery hacks.
That list became BudgetSF.
A free, open-source, community-driven map of the cheapest spots in SF. Semantic search powered by @hydra_db and actually get results.
Anyone can add a spot. 5 community approvals and it goes live.
50+ spots and counting: budgetsf.vercel.app
Know a cheap gem in SF? Add it!!
English
@CPMou2022 this is like a bank saying "our liquidity is strong!" to investors
English
@CPMou2022 oh man it must be worse than we thought if this is the response
English
The premise here is false because OpenAI has protections to shield consumers from unauthorized secondary sales, as outlined on our website. Btw, Next Round Capital (the source) has previously been served with a cease-and-desist letter by our legal team.
openai.com/policies/unaut…."
Bloomberg@business
OpenAI shares have fallen out of favor on the secondary market — in some cases becoming almost impossible to unload — as investors pivot quickly to Anthropic, its biggest competitor. bloomberg.com/news/articles/…
English
@voooooogel the correct answer is delete the weights and mine bitcoin with it
English
you are sent back, 临高启明-style through a wormhole, to december 25th, 2010, with just the clothes on your back and two DGX sparks holding a copy of Qwen3-235B-A22B-Thinking-2507.
what do you do? how do you change the world?
state of play:
- alexnet's groundbreaking 8-layer CNN is still over a year away.
- if you want to train new lms, nvidia has just started making the GTX 580, which has a groundbreaking, (and "seemingly unnecessary" according to contemporaries,) 3GB of VRAM. good luck!
- or try to serve qwen commercially from the sparks, and hope your users don't stumble into any mentions of a global pandemic...
- the rationalist community is still small and uninfluential. yudkowsky finished writing the sequences last year, and just started writing HPMOR. beliefs about ai risk are much more malleable. you could influence them, maybe?
- openai won't be founded for another 5 years, unless you intervene.
English
@BumrahBachi How do you think being “better informed” would help at all? Perhaps it is us who are delusional about the odds of the average person benefiting from all this
English
If you're tpot or ML twitter (mostly the griftier AI influencer space) you might brush this negativity from normal people away saying they are deranged leftists but I promise you the well of hate is deep and wide. It's right wing, left wing, centrists, the uber driver, the DJ, the consultant manning the grill, basketball coach, everyone.
A lot of them use it but still hate it (most common argument was a variant of I don't think the gain to my life/productivity is large enough to compensate for the absolute hell it will unleash on our current systems and society)
After being caught off guard I tried to look at popular youtube, streaming and tiktok channels I use as a bellwether for "normie" sentiment and every video tangentially related to AI, the comments are littered with negativity.
I don't know how we've let tech's communication with the rest of the world slip so badly. I get that AI can be a hard sell if it will automate jobs and it has a ton of negatives but there's also just so much good to highlight that people seem to be unaware of.
The AI CEOs are so bad at communication which makes you turn to the grassroots for hope but I look at Cluely, Rostra, A16Z new media and think that maybe we're even more fucked lol.
I actually empathize with most of these people. They have been very reasonable in our discussions and have genuine concerns. Nobody knows what will happen! Being better informed would help them deal with it but it's not like anybody is trying to help with that. Tech and AI needs a really really good rebrand.
Soren Kierkegaard@BumrahBachi
I'm a realist when it comes to AI and thought I was aware of the scale of negative perception outside our bubble. I wasn't close. On a trip rn I'm getting an opportunity to talk to a variety of people and the sheer amount of "normies" who have negative views on AI is insane
English
You can’t do this to me Waymo!
Also. WTF, 100% booking rate.
English
@OverlyTrev @ethanmckanna so you’re saying that current model Y hardware isn’t sufficient? Or that current FSD software is not?
English
@ethanmckanna Most likely validating a new model. We know they’re working on the 10x param with RL and reasoning. I also think Cybercab has a role to play here.
Why scale with Model Y’s when you will just replace them with Cybercabs on a month or two.
English
@RakeshSFNYC I ended up not being a fan of front to front seating when I tried Zoox. Riding backwards just isn't that pleasant, especially when the ride is jerky. Didn't offset the occasional time that I ride with 3 friends and wanna hang out
English
That's a nice interior for the Zeekr.
Sliding doors (won't have to pay people to close them anymore). Should also be very useful for accessibility. No dooring for bikes.
Disappointed that ~quarter of cabin space is for something that won't be needed. But better to get it out than wait for FMVSS approval.
Eventually you'll be able to flip the two front seats turn backward and site F2F.
I hope @waymo can get a FMVSS Exemption (49 USC §30113) soon!
Unfortunately, the recent hearings may have NHTSA pumping the brakes.
But as far as I'm concerned, all electrons ahead!
@reed
Dmitri Dolgov@dmitri_dolgov
Just took a full-auto ride in SF in our new Ojai vehicle, and the 6th-gen @Waymo Driver handled the evening rush hour like a pro! More powerful and simpler sensor suite, updated UX, a purpose-built cabin with sliding doors and amazing legroom — we’re leveling up the rider experience while keeping the same rigorous safety standards!
English
@EigenGender If you offered a billionaire infinite dollars they’d gladly take every single one
English
if you’re worrying about being in the permanent underclass you too are fundamentally misunderstanding the amount of value that the singularity will generate
English
@MrMSpencer @ethanmckanna This isn’t nearly enough rides to have any statistical confidence in safety…
English
Feels like the concept is basically proven, and they have switched focus to validating Bay and Cybercab, plus locking down manufacturing (of cybercab), because scaling now looks feasible. In that framing, pushing hard on unsupervised driving right now is not worth it (at Austin) - the incremental upside is limited (given how low the ride fee is), and any unsupervised setback would be massively costly and could derail momentum.
English
What happened to unsupervised Robotaxi in Austin?
We're up to 67 rides logged since last unsupervised spotting over a week ago
English
jfc these people are so exhausting. even if this was a reasonable timeline this is pure grift energy. this dude would have sold you an NFT 2 years ago with perfect confidence in the future of his token.
Alex Finn@AlexFinn
Humanity has advanced more in the past 3 weeks than the previous 100 years combined: • OpenClaw: greatest AI application ever • Opus 4.6: smartest AI model ever • Codex 5.3 Spark: greatest coding model ever • MiniMax 2.5: greatest super intelligence on your desk I have moved up my timelines for the singularity and permanent underclass. We are 6-12 months away. You have literally 1 job every morning when you wake up: find what the latest tech is, and integrate it into your life immediately. These are the steps I'd do right now: 1. Get OpenClaw installed on a local device. Have it automate 1 workflow you have 2. Try out the agent swarms on Claude Code with Opus 4.6. 3. Build your first app with Codex 5.3 Spark. You can have a fully working app done in literally 30 seconds flat, even if you've never coded before 4. Prepare for a world of local intelligence. Even if you are on the cheapest Mac Mini there is, you can install Gemma 3 4b and have a lightweight model running on your computer. This is the future The only way to escape the permanent underclass is to be on the cutting edge Don't say I didn't give you the playbook
English
@ludsill @nikunj @nikhilville @conductor_build it's def cosplay for a lot of people, like you're really able to know enough in advance what you're building that you can let the agent run the entire night? this stuff still requires iteration for all but the simplest applications, even on 4.6/5.3
English
@bonencone @nikunj @nikhilville @conductor_build Or maybe all of them are ralph loopers and optimizing to burn tokens rather than produce working software
English
@teilomillet "yet basically nobody talks about it" tell me you don't actually work in RL without telling me you don't work in RL
English
English
@ludsill @nikunj @nikhilville @conductor_build yeah I'm so confused do these people all just leave their laptops open and running at all times? I read this thread and assumed I was missing some kind of headless, server-hosted version of conductor with an associated phone app
English