u know damn well if julius was hallucinating input files like it did there ud make us issue a refund +drop everything to push an immediate mitigation
i’ll defend devin and their team here. most of the critique here are things you’d hear from a senior engineer (dinosaur) who’s watching over your shoulder and constantly negging you. it’s completely regarded
most of his arguments are strawman
- devin used an incorrect version of pytorch (duh, knowledge cutoff)
- “it took 36 minutes to do this task but took devin 6 hrs” (i think the fair comparison is how long it’d take an average upwork tasker to do this)
- devin could have read the README (instruction manual) and saved installation time with a command. (honestly who does that. any good engineer would just starts playing with the thing and poking holes instead of following the instruction manual word for word)
- code quality was nightmarish (no comment)
5
2
154
100,216
julius is basically entirely backend engineers bottlenecked by figuring out how to make a button render quickly in the right color and location and cannot understate how much we appretiate people who understand this stuff helping out
vercel developer support is god tier. emailed rauchg for some help, he replied within 19 minutes and looped in his team
7
1
124
52,695
family coding best coding
Current status: NYE coding marathon with @badphilosopher. Most relaxing part of the year.
There's a new version of this post
3
1
73
25,703
We built an interface around #GPT3 to create markket.ai, a tool to computationally produce text-based ads for products to different contexts. We're expanding our pool of limited beta users, check it out!
4
13
66
if anyone reading this is looking, we're hiring at @juliusai for scaling and securing sandboxes. sf startup, hit us up.
3
5
33
11,871
Big day for the team. Data Connectors are live, and they make it simple to go from "I wonder…" to "Here's the answer."
- Plug in Postgres, Google Drive, SharePoint
- Julius learns your schemas and context
- Ask questions, get verified insights in minutes
- Built with team controls and safety in mind
Huge thanks to our engineers and early customers who pushed us to make this both powerful and safe. Curious what you'll build with it!
Today: We’re launching Data Connectors 🚀
Connect directly to Postgres, Google Drive, SharePoint + more. Julius reasons through your data, learns your schemas and pulls insights in minutes.
3
4
35
4,812
wooo
We're excited to supercharge @juliusai's mathematical computations with our newest partnership...bringing Computational Intelligence to Artificial Intelligence 😊
Check them out here: julius.ai/
2
2
35
6,829
is good model
we have been testing GPT-5 for a few weeks and it quickly became my daily driver in Julius. it’s really good at coding and writing
live now for all users in @juliusai
3
1
32
5,104
in celebration, we finally launched the improved dark mode experience everyone's been asking for!
1 million data visualizations generated on Julius 📊🎉
@juliusai newest milestone
3
1
30
7,265
concerning that pmarca slowing down ai progress by seducing ai engineers from julius into hollywood
1
29
12,545
Formalizing something and giving it a name is still useful though. People were hitting the aidungeon scorebots years ago with prompt injection (aidungeon.medium.com/how-we-…) but no one realized it was interesting enough to codify it at the time.
1
26
17,949
we've started using 4.1-nano to read a bunch of chats quick to summarize findings / do meta analysis and provide links to them. is pretty cool.
1
2
25
4,039
i think i got in a few words between the 'um's
2/ JuliusAI
Jupyter notebook, on steroids.
Chat powered data analytics and AI agents, all in a notebook interface. Answer any question about your data with a single prompt.
@badphilosopher @0interestrates @juliusai
5
1
23
7,592
playing with env vars in julius so you can add keys for important tasks like letting gpt4 use the openai api to write and evaluate poems julius.ai/chat/d6987b8a-5fde…
2
1
23
21,092
now we have an excuse to be on our phones during the business meetings cause we're using julius to crunch the numbers
Introducing the Julius iOS App📱🤖
AI that:
- Solves Math
- Analyzes Data
- Creates Visualizations
- Writes & Executes Code
is now in your pocket 🚀. Download here: julius.ai/iphone
Android next!
3
3
23
7,433
dedicated to helping delve into data
2 years of building @juliusai and we’re just getting started. the world is drowning in data but starving for insights. julius empowers every curious mind to seek their own insights
i’m so excited for what’s next. we couldn’t be here without our amazing community of users ❤️
2
1
22
4,696
Text-Davinci-003 has some really cool properties when it comes to getting structured output and we figured we'd write about it because it's really useful engineering.hyperwriteai.com…
2
3
22
7,030
having a degree in journalism, always makes me happy to see julius helping people make sense of the data around them
worlds gonna keep getting more complicated and need a way to understand + convey meaning and significance
2M impressions. citizen journalist analyzes polling data using Julius to expose election fraud in eastern europe
analyst in your pocket
1
20
3,530
So we ran the numbers and it turns out GPT3's been able to do better than chance on Word in Context all along, and it's been improving with each new model.
engineering.hyperwriteai.com…
1
5
21
6,178
if you like playing with infra, apply now
Julius is growing quickly and hiring Infrastructure Engineers in SF!
if you’re excited about solving hard infra problems around AI + code execution and have experience working with Kubernetes, Docker and cloud platforms, pls reach out
DMs open and application link below
1
1
20
17,703
looking forward to a nice relaxed summer since it looks like all the ai providers aren't gonna be making any big changes for a while
2
20
1,818
underrated part of the holidays is you get to pull all nighters hunting down bugs and no one complains when you show up to the office at 3pm
21
1,065
google ads integration in julius (shoutout @heyaleksandr) might be my favorite thing we’ve shipped lately. can’t believe we were out here living like cavemen before this.
4
1
19
10,790
no idea how this sorta automation changes the internet, but this is a good team to be doing it
The next billion dollar company will be powered by Browserbase.
We already help hundreds of AI startups automate the web at scale.
Now, we've raised a $21 million Series A round, co-led by Kleiner Perkins and CRV, with participation from Okta Ventures.
What will you 🅱️uild?
1
1
18
2,474
these llms can get like 80, 90% of the way there for a lot of tasks but then turns into whackamole on the last bit (speed/cost as well). there's a lot of potential in making it easier to get the finishing touches in.
Introducing AI Graph Editing in Julius 📊📈
Users can now tweak and customize their graphs with AI and just natural language
Easily modify the plot legend, size and title or just tell the AI in simple english how you want it modified😇
Give it a try!
1
2
18
4,947
i heard of this newfangled thing called a neural network that's essentially a lossy compression algorithm but I don't know if anyone's figured out how to get them to do anything with text or images yet
16
data wants to be shared
Our biggest update so far: We're excited to announce Collaboration in @juliusai
Julius is now your team's AI Data Analyst with built-in realtime collaboration. Check it out below
5
3
17
2,852
i cant tell if theyre dunking on us but theres something nice about simplicity
Pro tip: Try our built-in @ Task smart chip for even easier task tracking → goo.gle/4adNCVd
1
16
4,713
Updated the WIC GPT3 benchmarks, looks like gpt-3.5-turbo does about the same as text-davinci-003 for JSON/Code formatted prompts, although slightly worse when text only
github.com/OthersideAI/WordI…
1
3
15
2,225
welcome to julius
After 2.5 wonderful years at @tryramp, I’ve joined @juliusai as Head of Marketing!
Julius is an AI-powered data analyst, making analysis accessible to all—no coding needed. Trusted by millions worldwide.
We’re growing fast and hiring! DM me or reach out at careers@julius.ai
1
15
3,454
schools adapting ai into their curriculums are going to be the norm
👨🏫 Professor: Use Julius to solve a financial problem and have a deliverable
👨🎓 MBA students: Used Julius Workflows to build a real estate investment risk app
Run your analysis and build financial apps on Julius? The possibilities are endless.
1
1
12
831
today i learned that .copy() in python is a shallow copy and u need to use copy.deepcopy() if u don't want to crash things when slightly mutating things
1
13
941
lists of stolen credit cards are cheap (card testing is now part of the background radiation of the internet unfortunately). paid isn't that great of a signal if at all when looking to prevent scam and fraud especially with a skeleton crew.
10
1,409
glad to see you guys digging more into the awesomeness of llms interacting with tabular data ❤️
9
1,476
kinda surprised how many people think josh gently expressing disappointment towards racism is somehow the spanish inquisition. y'all can do better like wtf.
10
2,857
So the API can help translate from British to American! One of the harder issues with content analysis.
2
1
10
it's hard to figure out what to hard code vs what to let the models figure out on ai driven products because the speed/cost constraints on random cool things just keep getting eliminated every few months
10
1,061
glad to have you
after two and a half years, we’re shutting down Ellis.
we built something we were proud of. we helped real people. we shipped constantly, learned even faster, and made it further than we probably had any right to. but we couldn’t grow fast enough. eventually, our runway ran out.
i still remember our first real sales call in the philz on polk street. all we had was a landing page and an idea. no frontend yet, just promises. when the Stripe balance ticked up from zero, we manually onboarded the customer on the spot. someone believed in us.
when our first investor check finally cleared, i used it to pay rent. we were that close.
for a long time, i thought if Ellis failed, it meant i failed. but that’s not true. building something from nothing is brutal and beautiful, and i’ll never regret taking the shot. i’m deeply grateful to everyone who supported us; users, teammates, investors, friends and especially family. thank you.
i mean that in the biggest way possible. your belief carried me further than you know.
there’s no tidy way to fail. it’s messy, quiet, and humbling. i’ve spent weeks replaying every tradeoff in my head. but deep down, i’m proud—because we really did it. we showed up and made something out of nothing. from the philz on polk and Shack15 to a Starbucks in El Paso and Aaron’s living room (my favorite office by far), we kept showing up.
now it’s time for the next chapter. i’ve joined an incredible team at julius, and it feels good to build again.
1
10
3,122
hard problem. so many times with word in context getting excited about improvement on first 100 examples and then worse performance on next 500.
ab test in prod gets better scale but still problematic =/
1
7
1,713
missing from this analysis: Honor
for many its a genuine value
How much was OpenAI really worth, a week ago? And what are the employees thinking about right now?
Here’s a thought:
If OpenAI has a lot of breakthrough, difficult to replicate IP (code, data, infrastructure etc), with a real business case, a lot of employees would stick around, rather than giving up the unvested part of their stake in something massive.
If it was never really worth what people thought, and everything could easily be replicated, everybody will bolt.
If they do, it’s a strong indication that generative AI has been overvalued.
1
1
8
1,423
i still don't know what self aware means in the context of gpt/claude, but having either power the awesome infinite self-reflecting while-loop chatbot from the scale hackathon last year works out of the box once i upgraded the apis for it :)
4
1
9
3,644
early test case for setting up notebook crons for julius to connect to a data source, create actionable insights, and then take action on them using connections. it's really fun to see the change history.
Excited to announce our Google Ads connector in Julius. Hear from @heyaleksandr on how you can get instant insights on all of your ad campaigns, spend, keywords and more directly in Julius. 📊
2
9
725
feel free to dm
Julius is hiring in SF 🌁 we need all the help we can get!
- frontend/fullstack engineers (react, typescript, nextjs)
- platform/infra engineers (kubernetes, terraform, gcp)
- marketing (content, growth)
- senior product design
for engineering: DM @badphilosopher
for marketing: DM @loclug
for design: DM me or @lukvsm
4
7
1,030
Claude can also do few shot word in context in JSON and Code! Got access at a hackathon over the weekend so added it to the list. It seems to do its best with JSON prompting. github.com/OthersideAI/WordI…
2
8
1,180
GPT's all about the logprobs.
GPT-3 is a super powerful tool, but there's still a lot of unknowns around how to use it effectively. Because of that we've decided to write a series of blog posts around how to use GPT-3 from what we've learned building AI Dungeon. Check it out here!
medium.com/@aidungeon/contro…
6
learning to code is fun (cool github viz by @sallar and friends github-contributions.vercel.…).
new years resolution: not have those blips next year.
2
7
1,260
Just replacing the from and putting in an @ gets most of the way there :P Thought draftjs was gonna be a bigger pain. I suppose should do cleanup after the search (and intercept clicks) but that's a lot of work. gist.github.com/matthewlouis…
1
7
There's still a lot of handwaving on best practices.
It's expensive to benchmark prompt usage and not centralized anywhere; would be amazing to have datasets and scoreboards using hold out test sets with free/subsidized compute for community to figure out best usage.
1
6
. @notmorejaden is a beast, really took lead on getting this knocked out over the winter break
Our first launch of 2025. Introducing Julius Teams.
You’ve asked for better ways to work with your teammates in Julius, today we’re making that possible.
👥 User Management
🔒 Privacy Controls
📊 Unlimited Insights
💳 Centralized Billing
Create your team today (link below)
6
593
* warning: if u run this, i don't think the off button works. it was a hackathon and didn't get that far.
github.com/matthewlouisbrock…
6
519
Making progress on getting GPT-3 to compare/summarize news coverage across sources. Can be a bit slow on the fly but it's not too bad for grabbing differences in coverage across platforms and doing content analysis without me having to read much. Still a ways to go though.
2
6
i agree and it applies to everyone - i doubt anyone has a very good understanding of why things are happening in the world
except for maybe the quants and then only in a very narrow sense
3
329
After playing around with how to format the prompt to walk itself through a task for a bit, figured worth playing with having a prompt double check its output as well. Gives improvements to arithmetic in GPT-J (and also 3 but J's cheaper :P)
towardsdatascience.com/1-1-3…
1
1
6
imagegen go brrr
Its official @prodialabs has the fastest image generation in the world, thanks to the teams at @ArtificialAnlys and @BlackForestLabs!!
1
5
686
Also ran GPT4 through ANLI - gets 69.75% without any major tricks. Gotta go back and figure out what was getting 40% on original Davinci again though because didn't get it with these runs. github.com/OthersideAI/ANLI
5
676
So just using TFIDF for context stuffing with GPT API gets 9% exact match on the Break benchmark. (Which will be an important step for chaining prompts back into one another I suspect).
leaderboard.allenai.org/brea…
5
needs to happen fast. students already simply paying humans to do their homework and those companies make their services cheaper by abusing ai free tiers. efficient marketplace and all
At the core of the AI crisis in education is that the system invents fake work for ppl to do so that they can be evaluated against the quality other people’s invented work.
As soon as the frame switches to *actually making progress*, the incentive is to use AI to the greatest extent that it’s useful, not to hide in fear of a future ai text classifier.
1
5
1,443
It was much harder before text-davinci-003, but also their safety policy made it a bit hard. We'd been doing our code interviews for @HyperWriteAI with chatbots using original davinci and then text-davinci-002 and they'd been making a lot of improvements over time.
5
551
It can get 67% on WiC test set for figuring out where people are using the same word in different ways, gptprompts.wikidot.com/lingu…. Not at human level (which would be 80%), but not random.
5
anyone else having a bunch of @auth0 issues starting about 20 minutes ago?
5
1,728
putting stable diffusion and various llms into every website on the internet as it should be
🎉 We have some BIG news for @HyperWriteAI users! 🎉
You can now generate amazing images on nearly ANY website with just a few words!
No more wasting time on Photoshop or searching for stock photos. Just let HyperWrite do the magic for you!
3
4
It's because of latency (we serve one). The LLM generates each token in sequence... and if you're doing hundreds tokens this can take several seconds especially when underprovisioned. By serving each token as available (just stream=True on the APIs) it shows it's working faster.
5
430
we're just getting back to 100% on the mandatory trainings, don't make us add financial literacy to the list
1
2
26
Similarly, we've been using pseudo-python prompts, where we declare a few classes and use their functions without defining them and let text-davinci-003 guess what the output would be (except overriding the print with f-strings so we get the output formatted the way we want).
I mostly write GPT-3 prompts in a formal subset of JavaScript consisting only of top-level assignments of JSON literals to variables, using a stop sequence of ";⏎" (semicolon, newline) so generations can be parsed as JSON.
Also, here's the wiki on Taken 4: The Musical:
5
997