Chase
@ChaseMc67
Followers
1,431
Following
1,548
Media
331
Statuses
3,467
Founding engineer @codegen | Prev @roblox , @intuit | all opinions are shitposts
San Francisco, CA
Joined February 2011
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Game 1
• 806507 Tweets
D-Day
• 737115 Tweets
Bannon
• 414280 Tweets
Pakistan
• 320154 Tweets
JUNGKOOK IS COMING
• 267874 Tweets
Celtics
• 202412 Tweets
$GME
• 133589 Tweets
sabrina
• 109432 Tweets
Dallas
• 108093 Tweets
Mavs
• 105063 Tweets
ENGLOT LoveBully EP1
• 104320 Tweets
#วันนี้3ทุ่ม15มีรักให้ร้าย
• 99926 Tweets
#NBAFinals
• 66808 Tweets
Kyrie
• 65454 Tweets
Porzingis
• 63785 Tweets
Tatum
• 52744 Tweets
barry
• 49540 Tweets
LINGORM FANMEET TICKET
• 40189 Tweets
ME CAIU TAO BEM MH E GM
• 39167 Tweets
Dr Pepper
• 37757 Tweets
Doncic
• 30677 Tweets
Alabama
• 29983 Tweets
Dr. Phil
• 24505 Tweets
Jaylen Brown
• 16386 Tweets
Kelvin
• 10381 Tweets
#スターレイル予告番組
• 10263 Tweets
Pinned Tweet
Incredibly excited to finally share publicly what we've been working on the last few months. Playing with this demo and product got me so excited I HAD to work on this. If you feel the same way, we're hiring 👀
Excited to share what I've been building 🚀
Introducing
@codegen
⚡ Agent-driven software development for enterprise codebases ⚡
We've raised $16mm from
@ThriveCapital
and others to bring this to the world -
👇 More below
67
123
1K
6
1
29
@trunarla
I can count on two hands the number of single, principle engineers I know, who make like 700k+/year but live in a $2000 appt and gripe about spending $5 for a coffee or staying at a non-budget hotel during their one vacation they take a year.
12
7
718
@mathemagic1an
If you’ve never heard of Alex the Parrot you should definitely look him up. He used to live in the MIT media lab, and they built him a little computer he loved to play with
5
26
652
@MegaBasedChad
As OAI gets closer to AGI it’s becoming more human. Soon it’ll be responding “I’ll get to it tomorrow”
10
8
660
@KevinNaughtonJr
you have a round old-timey clock (the kind with the minute and hour hands). It starts at 12:00.
I say to you “add 50 hours to the clock”, so you spin the hour hand around 4 times + 2 hours. It now says 2:00.
Now we give someone else the clock, and ask “how many hours did we
14
26
654
LLM prompt research papers in 2023 be like:
“GPT4 Reasoning Improvements After You Tell it a Lil Joke” - Deepmind, MIT, Stanford et al:
Telling GPT4 a lil joke increases reasoning ~25%. Knock knock jokes seem to work best. Here is a list of jokes we tried
19
37
536
PSA: apps don’t work on Vision Pro while walking. Everyone doing this is literally just swiping at air and pretending while wearing it
2
36
305
Can confirm gpt2-chatbot is definitely better at complex code manipulation tasks than Claude Opus or the latest GPT4
Did better on all the coding prompts we use to test new models
The vibes are deffs there 👀
11
11
274
Claude opus is so insanely OP.
You can just pass it a handful of 300+ line files, ask it for any change, and it spits em back out perfectly.
GPT-4 has significant issues spitting code files back out without making weird stray changes to comments or omitting code
19
11
254
@zack_overflow
You need to have this automatically enable after your WPM exceeds a certain amount in a session.
So it’ll just randomly turn on when you’re really in the zone, like a score multiplier in a video game
7
3
251
meanwhile there are still senior engineers out there who think LLMs can't write code because GPT3.5 made a typo when they tried, or they haven't even tried it at all yet...
Today we're excited to introduce Devin, the first AI software engineer.
Devin is the new state-of-the-art on the SWE-Bench coding benchmark, has successfully passed practical engineering interviews from leading AI companies, and has even completed real jobs on Upwork.
Devin is
5K
11K
46K
25
16
237
@jennsun
“You can just be my employee” is the new “we can just be friends”
I don’t see how anything could possibly go wrong with that
5
7
199
@dejavucoder
It’s literally just a group of the 10 most cracked out programmers in the world
2
3
195
@ronawang
We actually have human labor masquerading as AI in every timezone, thanks very much
1
2
157
1
7
151
@LucasPJohn
He’s gonna need to sleep for a month after this, if the stimulants don’t kill him
1
1
143
@Lola_lmao7
Listed on AirBnb as “Harry Potter theme room. 100% accurate to the movie. Live the Wizard experience”
0
0
138
@Suhail
They were going to, but bard didn’t get 100m users so they just scrapped the whole thing and re-orged the executive sponsor somewhere else
3
2
118
0
0
114
@Doffee_
As someone who used to literally work on “AI creating video games” at Roblox, I say, give it a year
2
0
115
@MKBHD
“I’ve been living in this building for about two weeks now, and I’ve got some thoughts”
0
0
108
@tszzl
I love that the Amex lounge in SFO has a lil roped off area for the black card holders, and it has crappier chairs than the rest of the place 😂
Literally a lil rich-person zoo exhibit for the rest of us
3
1
108
@JudgeMegapolis
@trunarla
Yes, but we’re talking decisions like giving up something you’d really enjoy/use every day so you can save an extra $2k/year when you’re already saving like 200k+
Decisions add up but you also gotta enjoy your life at least a little
1
0
99
@gfodor
Complex agentic flow ontop of GPT-4-32k that can turn GitHub issues into PRs. Still early but already working well enough it’s solving a huge chunk of our own tickets
6
3
79
@BogusThought
The fact that Elon continues to push as hard as he does, after all the success he's had, is the most alien part about him tbh. We're lucky to have such a messed up bro
0
2
77
@paindesuxd
When I worked at Roblox I had someone dox my phone number from my LinkedIn account and spam my phone with threats to unblock his Roblox account 💀
Never call/message someone’s personal number unless THEY gave it to you
6
1
70
@svpino
My new favorite dev interview question: Build a full stack app in a language you don’t know using ChatGPT.
Watching what questions they ask and how quickly they can learn something new is sooo much higher signal than a random leetcode
6
4
68
1
0
66
@mathemagic1an
One Alex story: researchers had a bird intelligence test for Alex to take. String attached to a perch with a treat dangling, bird needs to figure out to pull up the string to get the treat. Alex sees it, looks at the researcher, and says “hey you, pull up treat. I want”
1
5
62
@LatinxPutler
So true. I feel like the hardest part of really learning a new math/science concept is finding some practice exercises to practice and ingrain it
2
1
58
@gfodor
I actually would have been surprised if the US govt let OAI die without intervening at all.
Seems like it should be at the level of national security interests
3
0
54
@SawyerMerritt
If Elon really wanted to show how impenetrable the cybertruck is, he should leave it in SF with a backpack inside
6
3
53
@Austen
Netflix costs $9/month. No one person could possible provide more value through Twitter than all of Netflix provides. MAYBE a $10 subscription to ALL of Twitter, but definitely not an individual person
3
1
53
@tferriss
I use OneSec to disable all my dopamine apps for periods of time and it works pretty well
also a tripple-click-side-button shortcut to put my phone into grayscale so its less stimulating
5
1
54
@KevinNaughtonJr
Sounds like a process problem. Y’all need some GitHub actions and branch protection rules.
And also more competent coworkers
5
1
46
@mealreplacer
Did ChatGPT consumer the water? Or shoot it into space? Or was it just a loop with 1 gallon in it that they used 500 billion times?
2
0
49
@danshipper
Can confirm it does better than GPT4 or Claude Opus on all our hardest coding prompts
4
0
48
@itsMattMac
Some random model just popped up on lmsys, which is one of the primary sites people use to compare different models.
Unlike other models this one has unknown origin. It also has a weird name “gpt2-chatbot” and it seems to be INSANELY capable.
People think it might be openAI
2
2
45
@yacineMTB
Don’t you know the kidney trick? God gives everyone their startup capital when they’re born
your kidney can go for like $300k and you only need one to survive
god gave us all startup capital. you just have to want it bad enough
273
1K
9K
1
1
45
@Son_of_Slumber
@bryan_johnson
He did more research to figure out what specific things have the most benefit if you can’t do the whole thing
3
0
42
@benz145
Definitely worse.
PC VR has like 100x the friction or standalone, so adoption would likely be 1% of what it is now
2
0
44
@dr_cintas
Can confirm it does better than GPT4 or Claude Opus on all our hardest coding prompts
2
0
37
Codegen team startin to get kinda big 👀 can't even remember everyone's names anymore
4
2
38
@QuinnyPig
Try explaining to your mom what you do as a software engineer working on a 5+ year old product.
“But isn’t it done already? Why do they need you?”
3
0
37
@kyliebytes
If you convert each color to ascii, and order the hearts by the date the author joined openAI, it reads out “touch grass”
2
3
37
@tekbog
Start measuring developer velocity by commits and tickets
Write a script to turn any PR into 1-commit-per-line (or 1 PR per line)
Make a ticket for each commit
📈
1
1
36
@petergyang
Gemini probably isn’t that much better than GPT4, but they wanted it to look a LOT better.
Benchmarks can be gamed and don’t always represent real-world use cases super well, and even on benchmarks it’s barely better
4
2
36
@VCBrags
Every time I tell someone I went to school in Canada they say "oh wow, Waterloo? thats a great school"
and then I have to tell them I went to U Alberta.. and explain what "alberta" is.
1
0
33
@lee_schmalz
@KevinNaughtonJr
A shitload of clock arms and the the output isn’t just one number (the time) but a shitload of numbers. Theres a shitload of gears in between the arms and the output when you move one arm it changes many outputs. so it’s really hard to change the clock arms individually to get
1
0
31
First AI came for the interns, and I did not speak out—because I was not an intern.
Then AI came for the junior engineers, and I did not speak out—because I was not a junior engineer.
Then AI came for me—and there was no one left to speak for me.
4
3
30
@asimo3089
Can confirm they definitely care, and there are probably doesn’t of loooong slack threads about this right now trying to find a good solution
1
0
29
@petergyang
ChatGPT’s recall gets worse as number of tokens in the context increase.
Beyond a few k tokens it can’t reliably parrot back its entire input.
There was a good paper that had a graph of recall/num-tokens but I can’t find it ☹️
4
0
29
2
0
27