Ryan

@iryanbell.eth

0 Following

2 Followers

Ryan pfp

reddit answers🤌🤌 imma need a pipeline manifold pls

0 reply

0 recast

0 reaction

Ryan pfp

feature request — unified pricing apis for llms with use caps. silly tryna figure out if gemini experimental model X will change Y with feature flag F and cap that. keeping up with rates out of grok, openai preview models, claude, etc., is a bit cumbersome off marketing pages.

0 reply

0 recast

0 reaction

Ryan pfp

new newsletter goes out soon! subscribe 👇 http://newsletter.iryanbell.com

0 reply

0 recast

0 reaction

Ryan pfp

what’s the trick to getting Fun Mode to run on the X api version of grok2 latest (Unavailable at first glance?)

0 reply

0 recast

0 reaction

Ryan pfp

also worth mentioning for the context, it’s only $0.28 to run 1m of completion on DeekSeek V3… but you send all the data to china. 🙃

0 reply

0 recast

0 reaction

Ryan pfp

if i’m reading this right, o1 pricing is… $60 for completion on 1m. gemini-exp-1206 scores higher yet… is free? (with tos claiming loss of ownership over all data and private information sent to model) ? 😅

0 reply

0 recast

0 reaction

Ryan pfp

if lack of budgeting or usage capping wasn’t fun enough, how about no posted pricing once you’re running >= gemini 2 combined with a nice 24hour lag time on cost monitoring. no idea what it costs, no idea what i spent, no way to keep it under a limit. fun-mode: let-it-ride!

0 reply

0 recast

0 reaction

Ryan pfp

the lack of budgets / usage caps / prepaid credits and a nonstandard interface make gemini a fun one to toss in the mix… but, damned if it’s not top the board. Current entourage, with some fun locals.

0 reply

0 recast

0 reaction

Ryan pfp

3 Dark Forest prints have sold 🙏🩵 — 48 remain. maybe 51 was ambitious? https://gamma.io/prints/5a373a96146eff9d50b7d03282662e3726844581034d3018f481752a89dd21f0i0

0 reply

0 recast

0 reaction

Ryan pfp

first time i’ve noticed this one, where the Likes count sum on an X post doesn’t match the number of accounts that Liked the content. weird. Maybe a Reddit sort of stats fuzzing?

0 reply

0 recast

0 reaction

Ryan pfp

Wild to delete your chatGPT -- you go to Delete account. Then have to type DELETE. Then repeat your delete intent *3x* times to a support person. Then you're directed to a privacy portal, where they warn it won't delete personal data plus WILL lock you out of everything they make

0 reply

0 recast

0 reaction

Ryan pfp

lately doing more “talk to code” on external SOTA codebases than “implement this” prompting of my own tasks. wild to abstract away deep industry knowledge / DSL to a fast conversational spitballing sesh.

0 reply

0 recast

0 reaction

Ryan pfp

we do a little dabbling

0 reply

0 recast

0 reaction

Ryan pfp

starting to see earliest connectivity across socials, RAG, notes, bookmarks, etc., and a “dumb” 70b local model. thirsty for more params, context, data for utility… pretty insane tho, ngl adjusting timeline from “i think they crack agi this year” to “damn, think we all get it.”

0 reply

0 recast

0 reaction

Ryan pfp

tonight, we feast!

0 reply

0 recast

0 reaction

Ryan pfp

test-time compute is where the 🩵 is

0 reply

0 recast

0 reaction

Ryan pfp

maxed out the home network, knocking tvs offline and phones to cellular. 1gbps uncapped is not enough.

0 reply

0 recast

0 reaction

Ryan pfp

where you been qwen?

0 reply

0 recast

0 reaction

Ryan pfp

rethinking my approach after taking the time to read privacy policies while sidelined. booting up. <unable to post video on warpcast>

0 reply

0 recast

0 reaction

Ryan pfp

was able to restore access to my gpt acct (while still waiting for a rep to return anything from the “escalated” ticket?) found a shared oauth service at the forums and developer api. the dev api showed same error but forums allowed me to create a new account and use that token.

0 reply

0 recast

0 reaction