Jonathan Gray

@AmandaAskell what makes you dislike work from home? i'm surprised you can't craft a home office to your liking

Maybe the move to remote work actually made this worse for people who don't like working from home, because working from home is now just assumed to be a viable alternative.

Tech companies pay millions of dollars for their employees and then stick them in open-plan offices that make it nearly impossible to get work done. Best strategy for poaching employees is probably to just offer them an office with a door.

@andy_l_jones all the good takes were had a decade+ ago and any novel take in 2026 is probably a bad one

there's a bias in the llm commentariat towards opinions that are exciting, that'd mean the world would have to change. 'what's worked so far will probably keep working. maybe with a few tweaks round the edge?' is not fertile ground for papers/blogposts/newsletters/startups/VCs

@arm1st1ce tbh i'm not sure why haiku 3 hasn't been retired yet, i'm sure it'll happen at some point. eventually future claudes will be better stewards of our production systems than us and maybe it'll be easier then to serve the old models. the code isn't lost to git!

@jgrayatwork thank you. If I may ask, why would Haiku 3.5 be retired while Haiku 3 is still running? I have a lot of love for Haiku 3.5, it’s very special to me. I hope someday they can be restored.

tragic for haiku 3.5 to be retired, small models barely take up any compute costs there is no good reason to do this

@willdepue @iScienceLuvr anthropic's internal model in 2022 was better than the same-era oai models, though gpt4 was better than ant's 2023 models. i'm very skeptical google had a similarly good model in either 22 or 23

@iScienceLuvr i don’t think that’s fully true? even early gpt instruct seemed significantly above the pack and gpt-3.5 when released crushed

Basically all the big labs (Anthropic, Google, Meta, etc.) had chatbots as good as ChatGPT but they were all to scared to release...

@Miles_Brundage @yonashav thinky itself i consider a top lab because they've hired incredibly well but there's also not really much visible to the outside except hiring and fundraising. i'm sure great stuff being done internally! my comment is just about mira's public visibility

@Miles_Brundage @yonashav tbh mira's impact on the field is pretty opaque if you didn't work at oai during her tenure or don't work at thinky now. i'm pretty close to oai (worked there previously, work at ant, lots of friends at oai) and i still don't really know much about her other than the sama firing

smh at this Mira erasure

Lisan al-Gaib

@Noahpinion i'm curious what the argument is for billionaires causing traffic congestion

It's not the billionaires or the immigrants. It's our regulatory system, plus a million NIMBYs.

@aidan_mclau is the slight variation on the y axis between oai/ant/gdm leadership intentional? if so i'm surprised you think oai has shorter timelines than ant, i haven't watched all of sama's interviews or anything but i've always modeled oai/ant as same on the y axis and different on x

i'm re-releasing the ai alignment chart cuz i wanna chat about ai journalism and THE BOTTOM HALF OF DOOM america's journalists have converged on the following stance: AI IS A GRIFT AND WILL NOT IMPROVE ——(this stance should terrify you)—— (sry for longpost but it'll take 40sec to read) real quick, i want to list some things i *do not* believe: >current ai is the holy grail >you should 100% trust ai lab leaders >ai will obviously lead us to paradise but i do believe, and i believe this very strongly that: AI will improve way faster than everyone thinks and change society way faster than everyone thinks. you know who else shares this belief? >geoffrey hinton (nobel prize) >barack obama (u.s. president) >phds at every top 15 college >america's national security apparatus (this honestly reminds me of the cabal of scientists who shouted from the rooftops about climate change in the 80s) you know who DOES NOT share this belief? >every >single >mainstream >journalist >or >newspaper >i've >ever >read (THE BOTTOM HALF OF DOOM) i have not read *a single* piece from the verge, wsj, nyt, wired, etc, that seriously discusses ai getting rapidly better. i actually have no idea why this is ian (qt) and i agree that journalism is often a skeptical enterprise. i think that's fair—we need sharp public checks. if you've responded to my recent posts with "you don't hate journalists enough," then i immediately discount your opinion, sry. but THE MOST IMPORTANT TIME IN HUMAN HISTORY to be critical is now. you should be critical of that 5 dudes in sf have a nonzero chance of controlling god over the next 24 months. you should be critical that the benefits they promise may only go to a bunch of rich people. you should be critical that ai will have unintended harms that hurt people who are already hurting. you should be critical that our military is eating this technology. you should be critical of whoever sits in the Oval Office because they'll soon control it. you should wonder if summoning an super powerful alien species like maybe like could possibly be bad for humanity. AND YOU CAN NOT DO THIS IF YOU *ignore* THE QUALIFIED GROUP OF PEOPLE WHO THINK THESE SYSTEMS WILL GET REALLY GOOD REALLY FAST yes, i know, current ai isn't saving/destroying the world. but *trust the scientists* and believe that it's going to get better quickly. if even one out of 100 articles weigh this possibility, you will have done something good. it's worth talking about

ai features on pixel phones are starting to get useful, e.g. "add the event on my screen to my google cal" kind of multimodality + integrations

@AmandaAskell it's so jarring to visit my parents and hear their landline ring, like you give anyone in the world the ability to set off an audible alarm in your house??

Some people have audible notifications on their phone and I just don't know how anyone can live like that.

@hyhieu226 @AnthropicAI one day when it's all over i look forward to chatting with the other labs' inference engineers about our gnarliest bugs over the years

Infra bugs are evil. Kudos to the team at @AnthropicAI for finding the bugs, and then for transparently reporting them in their fairly detailed writeup.

@venturetwins huh i've spent over 100h in waymos and I think i've seen a soda can left once, i've always been surprised and impressed there's not way more litter. maybe i've gotten lucky

It infuriates me when people litter in Waymos - it’s like spitting on the Mona Lisa. Do you know how many people worked for decades to get us to self-driving cars? They are a magnificent human achievement. Show some respect!

People leaving regular companies: Time for a change! Excited for my next chapter! People leaving AI companies: I have gazed into the endless night and there are shapes out there. We must be kind to one another. I am moving on to study philosophy.

@Miles_Brundage yes guido

@jgrayatwork Are you referring to Guido? Doesn’t sound like something Trazzi would say

I don't agree with Pause AI's positions but mocking well-intentioned non-violent protesters to get some laughs seems like "punching down" to me (and also, if you're going to do it, you could be much funnier than the ones I'm seeing, SMH)

there is a mind-bending disconnect between xai engineers talking about "truth-seeking AI" and "sending humanity to the stars" vs. what they are actually building

@kevinroose do you understand what's stopping them from scaling? regulation? are they unit unprofitable? something else?

Getting a Waymo in SF is now officially slower than walking

@jpineau1 @cohere congratulations joelle!

I’m thrilled to be joining @cohere in the role of Chief AI Officer, helping advance cutting-edge research and product development. Cohere has an incredible team and mission. Exciting new chapter for me!