_______ __ _______
| | |.---.-..----.| |--..-----..----. | | |.-----..--.--.--..-----.
| || _ || __|| < | -__|| _| | || -__|| | | ||__ --|
|___|___||___._||____||__|__||_____||__| |__|____||_____||________||_____|
on Gopher (inofficial)
URI Visit Hacker News on the Web
COMMENT PAGE FOR:
URI Show HN: Erdos â open-source, AI data science IDE
agnosticmantis wrote 16 hours 28 min ago:
This looks very cool, Iâm gonna try it later today.
Out of curiosity, why the name Erdos? AFAIK Erdos was neither a
statistician, data scientist nor AI researcher.
He sure solved many probability/combinatorics problems and famously had
many many collaborators.
jorgeoguerra wrote 16 hours 9 min ago:
No specific reason. Mainly because he was one of the most productive
and collaborative mathematicians of all time. We actually considered
"Poisson" at some point but ended up going with Erdos.
anigbrowl wrote 17 hours 31 min ago:
Apple Silicon only, might be worth mentioning on the download link.
dartharva wrote 16 hours 10 min ago:
I'm seeing a Windows download link?
jorgeoguerra wrote 15 hours 59 min ago:
The download button on the erdos/ page is OS specific, but you can
also find all the download links in the download-erdos/ page.
jorgeoguerra wrote 17 hours 12 min ago:
Thanks for pointing that out - will fix it asap
sosodev wrote 18 hours 31 min ago:
Does it support OpenRouter? I tried configuring OpenRouter as a "local
model" but it seems to silently fail.
WillNickols wrote 18 hours 8 min ago:
Not yet - we need to change the header configuration for that to work
(versus connecting to local models), but we'll have it available
soon.
thom wrote 21 hours 30 min ago:
Give me this, but with a very efficient, opinionated path to put models
into production. Give me accessible PM and customer friendly
documentation about features and model choices at every stage. Make it
reusable and easy to modify. Make it robust and scalable at inference
time, with metrics and dashboards tracking performance over time. This
seems like optimising the bit that's already fun, but I see a lot of
value in hand-holding a department through all the stodgy boring bits
and getting high quality analysis repeatably into customer hands.
mkl wrote 21 hours 33 min ago:
The choice of name seems pretty bizarre. The famous Erdos [1] was a
mathematician, not data scientist, computer scientist, or statistician.
URI [1]: https://en.wikipedia.org/wiki/Paul_Erd%C5%91s
jorgeoguerra wrote 21 hours 28 min ago:
Erdos is also widely considered as the most prolific and productive
mathematician of all time (in terms of publications and
collaborations). Hopefully you can be as productive with Erdos :)
mkl wrote 14 hours 51 min ago:
But productive with it in a different field from the person it's
named after? That's weird. It seems disrespectful to him to name
a product after him when its purpose is pretty much unrelated to
his work.
bigmadshoe wrote 21 hours 31 min ago:
He did contribute to/utilize probability theory. He came up during my
undergrad probability class because of this:
URI [1]: https://en.wikipedia.org/wiki/Probabilistic_method
puppycodes wrote 22 hours 5 min ago:
Looks interesting but i'm unclear what makes it "more accurate"?
jorgeoguerra wrote 21 hours 56 min ago:
When models edit the raw JSON behind a Jupyter notebook, they often
mess up the cell structure by adding extra cells, misaligning code,
or making bad edits. We fix this by giving the model the notebook in
Jupytext format instead, which tends to make its edits cleaner and
more accurate.
buppermint wrote 22 hours 34 min ago:
Very cool. Any plans to add support for local models? This has what has
prevented us from adopting Positron so far. We have sensitive data and
sending to third party APIs is not an option (regardless of their
stated retention policies).
jorgeoguerra wrote 22 hours 33 min ago:
Yeah, we just added support for local models. As I mentioned in an
earlier comment, if you have a local model with an OpenAI-compatible
v1/chat/completions endpoint (most local models have this option),
you can route Erdos to use it in the Erdos AI settings.
johannesf wrote 22 hours 57 min ago:
Have you done any fine-tuning or prompt-customization for the
R-specific work? I've found the models worse on R when compared to
Python, especially for more complex tasks. This looks cool, thanks for
sharing!
WillNickols wrote 22 hours 53 min ago:
Nothing R specific. In my experience, Claude is pretty good about
using tidyverse for everything. What was is flopping on for you? Our
thought on not fine tuning models is that whatever comes out in 6
months is just going to be better than whatever we fine tuned.
mritchie712 wrote 23 hours 31 min ago:
We started with a product like this at Definite ( [1] ), but it became
clear there weren't enough people willing to spend real money on a
product like it when Cursor / VS Code already have good coverage on
data science.
URI [1]: https://www.definite.app/
rubenvanwyk wrote 11 hours 21 min ago:
Not sure if self-promoting on every single analytics- or data-related
thread is in line with the ethos of HN: "Please don't use HN
primarily for promotion."
vednig wrote 23 hours 33 min ago:
I see Google acquiring Iotas in the future, that's how good it gets
harvey9 wrote 1 day ago:
Do you have the option to run on a local model? Lots of firms don't
want data or prompts going outside the local network
jorgeoguerra wrote 1 day ago:
Yep â if you have a local model with an OpenAI-compatible
v1/chat/completions endpoint (most local models have this option),
you can route Erdos to use it in the Erdos AI settings.
shuwan wrote 1 day ago:
I think Rao is more appealing to me since Positron already has that
kind of integration, while RStudio doesnât. Plus, Posit probably
wonât ever add an AI Chat feature to RStudio anyway.
WillNickols wrote 1 day ago:
FWIW there's a bunch of stuff Erdos has that Positron doesn't
(including having solved Positron's top 5 open GitHub issues):
1. Remote development via SSH or containers
2. AI that can connect to ChatGPT, local models, or our backend
3. In-line code execution for Qmd/Rmd files
4. Julia as a first class citizen
5. Multi-agent chats: as many AI sessions as you want and theyâll
all run in parallel
6. Windows ARM64 builds
7. Open source AGPLv3 license
8. A bunch of other misc items including read-write data explorer for
CSVs and TSVs, plots history sorted by file and time, searchable
help, a command history tab, etc
Maybe the biggest difference going forward is that Positron was a ~2
year dev project, whereas Erdos reached feature parity (plus or minus
some features) in about ~2 months and is now adding substantial brand
new functionality every week.
shuwan wrote 21 hours 13 min ago:
Will, thanks for the explanation. This changes my view a lot. Will
give it a try.
SamTinnerholm wrote 1 day ago:
I can't tell how this differs to Cursor from your website. How is it
different?
WillNickols wrote 1 day ago:
A bunch of specific things below, but the main point is that it
integrates a bunch of features that data scientists use that don't
come with Cursor.
Specifics (mostly reproduced from above):
1. R/Python/Julia consoles accessible by the user and AI
2. Optimized jupytext system for editing notebooks efficiently
3. Plots pane for viewing and tracking plots
4. Databases pane for managing SQL/FTP connections
5. Environment pane for managing Python/R/Julia packages and
environments
6. Help pane for documentation
7. An AI that interacts with all of that.
8. Open source AGPLv3
For me, the biggest difference in the AI usage is that the AI doesn't
need to write one-off python scripts for everything and run them from
the terminal because it can just use the console directly.
Centigonal wrote 1 day ago:
This is a good idea, although IMO source control, compute, and MLOps
integration are bigger but less flashy pain points for data scientists
than AI in notebooks.
If you're going to market Erdos as open source, then IMO there should
be a github link somewhere on your website.
mscbuck wrote 3 hours 15 min ago:
Will echo that one thing that would prevent me from trying this is
def the source control. Otherwise it does look pretty slick!
WillNickols wrote 1 day ago:
Thanks for the suggestions - we'll definitely add those to the dev
list. Also, the GitHub is [1] (and it's on the download page but a
bit small).
URI [1]: https://github.com/lotas-ai/erdos
DIR <- back to front page