Continuum
@UseContinuum
Followers
427
Following
4
Media
18
Statuses
50
Explore trending content on Musk Viewer
rose
• 692604 Tweets
bruno mars
• 403430 Tweets
Wizkid
• 236426 Tweets
Flamengo
• 119108 Tweets
Corinthians
• 110600 Tweets
Yankees
• 84496 Tweets
#GHGala7
• 75374 Tweets
Saints
• 58326 Tweets
Al Smith
• 50092 Tweets
Klaus
• 48031 Tweets
Noel
• 34337 Tweets
Allan
• 34209 Tweets
Maracanã
• 33772 Tweets
Cleveland
• 33458 Tweets
Guardians
• 32104 Tweets
Broncos
• 28149 Tweets
対象作品
• 22814 Tweets
Hulu
• 21959 Tweets
#911onABC
• 21872 Tweets
Playoff
• 20770 Tweets
DEPAY
• 20220 Tweets
Boone
• 17814 Tweets
Stanton
• 17732 Tweets
Rossi
• 17268 Tweets
Alcaraz
• 17242 Tweets
#ForTheLand
• 15519 Tweets
Ganso
• 15425 Tweets
Wesley
• 13876 Tweets
Garro
• 13209 Tweets
Yuri Alberto
• 13102 Tweets
Athletico
• 12521 Tweets
David Fry
• 12357 Tweets
Big Christmas
• 11876 Tweets
Weaver
• 11852 Tweets
#RepBX
• 11403 Tweets
Bo Nix
• 10660 Tweets
Pinned Tweet
Continuum is now generally available. Let your software explain itself:
8
11
132
Yeah so turns out there's still no way to hold an infinite amount of information in your head for free
3
1
52
Check out our fucking badass new landing page. The mesh returns at
4
2
41
It's so confusing to me how powerful and yet underdeveloped GPT function calling is. Like this thing was fine tuned to use tools at a cost of probably hundreds of millions, but it can't guarantee valid JSON output and there's no first party methods for counting token usage
6
1
41
Get detailed explanations and diagrams of your software architecture and data flow
1
0
36
Every other day a new paper comes out that claims to solve context window limits or RAG with some fancy sounding technique, and then when you actually read the paper it's just 2000 lines of Python doing recursive summarization in a slightly different way
2
2
19
🚢 New Continuum version🚢
✔ Sleek new UI (shoutout
@dankuntz
)
✔ Code knowledge graphs for Python projects
✔ GPT-4 Turbo with 128k context window
Excited to see what you all build with it!
1
3
15
Since the dawn of the field, software developers have been on a quest to automate themselves.
0
1
10
Interesting, but seems like a fundamentally worse version of chain-of-thought, since that will also give extra "intermediate" tokens but can provide information beyond the null pause token.
Also wow it's still so crazy how big the gains from CoT are, what a cool paper
Think before you speak: Training Language Models With Pause Tokens
- Performing training and inference on LMs with a learnable pause token appended to the input prefix
- Gains on 8 tasks, e,g, +18% on SQuAD
16
176
925
1
1
9
So many problems are provably undecidable and yet you gotta just find partial solutions anyway
0
0
8
@ChaseMc67
This is a neo4j (open source graph db) visualization of some graph structures we've been extracting from code.
1
0
6
@Suhail
It's when you chain 20 endpoints with a 5% chance of failure and get something that only works 35% of the time
1
0
6
The DSP framework also had some extensive work on using LLMs to generate their own examples:
Large Language Models as Analogical Reasoners
paper page:
Chain-of-thought (CoT) prompting for language models demonstrates impressive performance across reasoning tasks, but typically needs labeled exemplars of the reasoning process. In this work, we
1
95
362
0
0
4
It feels intuitive now how much prompting can affect performance (you gotta tickle the latent space just right) but it's insane to get a 3x solve rate just by kind of asking nicely
0
0
5
@abacaj
Check out the github readme brother
Yeah so turns out there's still no way to hold an infinite amount of information in your head for free
3
1
52
1
0
4
@amusing553
@tenobrus
Yeah we've managed to find that right balance too, but even a little too much mention of function calling in the prompts can lead to it totally ignoring the arguments list. It's fascinating.
1
0
4
@IgorBrigadir
Don't worry we're great at helping you produce useful code while incredibly sleep deprived
1
0
4
@tomchapin
Handling the complexity of enterprise software is the highest ROI task in the field at the moment, and LLMs are just now reaching that tier of capability. Try us out and let us know what you think!
0
0
2