MS.

Posts on X before Mar 16, 2024, 5:36:48 AM UTC

On X

@michaelstolarz ↗
@MichaelStolarz
Past 24hrs tweets from OpenAI employees: @gdb there's a simple joy in staying up late to finish one last feature @leopoldasch One year since GPT-4 release. Hope you all enjoyed some time to relax; it’ll have been the slowest 12 months of AI progress for quite some time to come.
@MichaelStolarz
the only real competition is claude 3 opus. mistral large would only be good if it was open source, which it's not. tbh, i dont think openai is sweating at all because... well, nothing has beat them by a landslide yet. and im sure their next release will put them ahead a ton.
@balajis
Suddenly, GPT4 has four competitors. - Gemini 1.5 - Mistral Large - Claude 3 Opus - Inflection-2.5 Links below. All are impressive. I like Claude the best. simonwillison.net/2024/Mar/8/gpt…
https://simonwillison.net/2024/Mar/8/gpt-4-barrier/
@MichaelStolarz
big if true. i signed up on the waitlist.
@maisaAI_
Introducing Maisa KPU: The next leap in AI reasoning capabilities. The Knowledge Processing Unit is a Reasoning System for LLMs that leverages all their reasoning power and overcomes their intrinsic limitations.
Media attached to this post
@MichaelStolarz
i've come to the realization that i'm doing just fine with claude 3 opus. i need to stop wasting time stressing about WEN GPT 5 and more time building with the already amazing AI tools out there that way, i'm even more prepared to utilize gpt 5 effectively
@MichaelStolarz
you literally cannot be wrong if you just, at the beginning of every week, tweet out "it's going to be a big week in AI". or in any field for that matter you're welcome, now you're a leaker
@MichaelStolarz
>has vision >cheaper >almost twice as better at coding open ai has to release something soon; their gpt 3.5 turbo is now outdated, so is their gpt 4 turbo (opus)... maybe pi day is the day. hopefully they release... something.
@AndrewCurran_
Haiku is coming for our old friend GPT-3.5 x.com/AnthropicAI/st…
Media attached to this post
@MichaelStolarz
accelerate MORE. FASTER.
@adcock_brett
OpenAI + Figure conversations with humans, on end-to-end neural networks: → OpenAI is providing visual reasoning & language understanding → Figure's neural networks are delivering fast, low level, dexterous robot actions (thread below) pic.x.com/trOV2xBoax
Media attached to this post
@MichaelStolarz
hasn't happened once since i switched to claude 3 opus btw
@krishnanrohit
This is going to be OpenAI's undoing. Not okay to brainwash your genius LLM into acting like a consultant.
Media attached to this post
@MichaelStolarz
this is a joke post right
@JD_2020
Dear X Algorithm & peers in the AI space — I’m writing today as I’m seeing many of you talking about @cognition_labs, who appear to have come out of nowhere, doing something @WeGPTai has been doing (and documenting with high quality content posts on X) for 10 months. In just fou… MoreLess
Dear X Algorithm & peers in the AI space — I’m writing today as I’m seeing many of you talking about @cognition_labs, who appear to have come out of nowhere, doing something @WeGPTai has been doing (and documenting with high quality content posts on X) for 10 months. In just four hours time, they’ve amassed more views than all our demonstration videos combined over the past year. Devin doesn’t do as good a job as us. And our demo is available for anybody to use, today, for free. But, our reach seems hard capped and theirs seems to be a significant subset of the entire engaged AI audience on the whole platform. What gives? This is absolutely coming from frustration right now. I’ve gone ahead and made a video to capture some of that emotion 👇 Cc: @elonmusk @karpathy @ashleevance @kateirwin @KettlebellDan @business @SeanMaggard
Media attached to this post
@MichaelStolarz
o no husbant, u jus finish coding bootcamp but now ur useress because of ai devin programmer, now we are homeress
Media attached to this post
@MichaelStolarz
if you're creating software from scratch, most templates are total ass and you're better off just... well, starting from scratch. software templates give the illusion of saving time but in reality you waste much more debugging someone else's code
@MichaelStolarz
the biggest problem with people thinking that AI is conscious is that literally nothing happens if you don't prompt it like, once it's done with a task, it's... done. until you prompt it again. yes, you can set it to re-learn or re-run or re-whatever but... that's still a task.
@MichaelStolarz
personally it being "better at coding and reasoning" is earth-shattering enough for me already doing so much with claude 3 opus, before that with gpt-4. hopefully gpt-5 is as cracked as everyone is saying

Quoted post unavailable

@MichaelStolarz
the way we talk about context windows today is similar to how people talk about floppy disks / how limited they were with storage options "back in the day" except "back in the day" now means a few years ago, not a few decades ago
@MichaelStolarz
my personal coding experience using claude 3 opus, I'd say it's 20% better. I 0-shot correct answers way more often than with gpt-4. but here are some aider benchmarks for those interested
Media attached to this post
@MichaelStolarz
(since mr. schiff doesn't follow me yet) question: where did you get this tuesday date for model release? i haven't seen any new whispers about this
@SpencerKSchiff
I had tickets for a Dune showing on Tuesday at 12:45 Pacific Time. I just switched them to 4:15 due to the potential model release
@MichaelStolarz
so i've been working on this readwise side project the past 2 days... come to find out, their API does NOT allow for full-text pulls, only summary/highlights/notes defeats the entire purpose of having articles/tweets/etc all in my obsidian brain with 1 click @readwise pls fix :D
Media attached to this post
@MichaelStolarz
this is why i'm starting to think that allowing too much "customization" for a user is a waste of time/overcomplicates things for the user yeah, settings seem cool to power users, but most users... aren't power users. they just want to click and be done
@dejavucoder
users dont know what they want. good products reveal what they want. great products reveal what they need.
@MichaelStolarz
gemini pro 1.5 is the first LLM to take my ~20 minute video raw transcript and give me EXACT, PERFECT chapters for it. no other LLM has been able to do so, not even claude 3 opus, especially not gpt-4
Media attached to this post
@MichaelStolarz
this upcoming thursday... well, it's a thursday. but it's also the 14th. march 14th was when gpt-4 was released something big this upcoming thursday.
@MichaelStolarz
They really need to either increase daily limits on Claude 3 Opus or let me buy additional chats or something
@MichaelStolarz
bruh
@OpenAI
You can now add extra security to your ChatGPT and API accounts by turning on multi-factor authentication. Set it up in account > settings.
GIF attached to this post