One improvement of the first "super-naive" approach is to break down the state into a whole hierarchy of files, rather than a single file. This helps reduce (but not eliminate) conflicts when multiple clients are making changes to different parts of the state.
No, I tested the paid GPT-4 last year on similar questions (animal cognition) and it was so bad I decided it was a waste of money. I actually don't care if it's maybe gotten better in the past year, and I'm certainly not spending money to find out. Last I checked the best LLMs still have a 5-15% confabulation rate on simple document summarization. In 2023 GPT-4 had a ~75% confabulation rate on animal cognition questions, but even 5% is not reliable enough for me to want to use it.
The high school AI tutor probably wasn't using GPT-4, but the district definitely paid a lot of money for the software.
I also hate this entire argument, that AI confabulations don't matter for free products. Unreliable software like GPT-4o shouldn't be widely released to the public as a cool new tech product, and certainly not handed out for free.
Humans have been doing that for years. The AI problem is so prevalent because it seems to put a magnifying lense up to the worst portions of ourselves, namely how we process information and deceive each other. As it turns out, liars and cheats tend to build more liars and cheats, also known as "garbage in, garbage out," which leaves me scratching my head as to what anyone thought was going to happen as LLMs got more powerful. Seems like many are afraid to have that conversation, though.
I have tried some chemistry problems on the latest models and they still get simple math wrong (mess up conversion between micro and milligrams for example) unless you tell them to think carefully.
How is it different from CSVFiddle (which is also based on DuckDB-wasm, I think)? Ability to have multiple tables loaded and running join queries against them would be crucial.
- You don't need SQL to transform and manipulate data in Pretzel - we have visual blocks where you can construct a transform chain (though we do support manual SQL blocks now!)
- We have better CSV parsing (I spent a fair bit of time on this!). See one of our test case CSVs here that breaks (https://imgur.com/a/O8XMfET)
- We allow plotting!
- We support PRQL - which IMO is a better way to build data pipelines (but we also support SQL)
- We have AI Blocks! You can use your API keys (or our server) to get SQL from text which you can edit
- We're currently adding support for Python via Pyodide (pyodide.org) so you can do complex data transforms, even train simple ML models right inside the browser!
- Lastly, we'll be adding a ton more features to Pretzel over the coming months (many of them have been asked for in this thread actually!)
Support for multiple files is crucial - we'll be adding that soon. Right now, we're rethinking our whole task execution architecture so any major features will have to wait until then.
AgentHub was open source when I first started it. I was really excited about the idea of people building their own integrations and fostering some sort of community around cool automations.
We noticed a few things though. 1. People who were most excited about no code did not want to contribute code to the project.
2. We were 95% open source because we were dealing with credentials and sensitive info on our hosted servers. This 5% of obfuscation was enough to make contributing annoying since you didn't totally understand how some pieces fit together.
3. We were adding new features and redesigning aspects of the system so often that it felt simpler to close it and accelerate. Features like node versioning and secure credential storage made it all quite difficult to maintain in an open way.
I do still love the idea though. Having people contribute their own integrations would be an absolute dream.
Flowise https://flowiseai.com/ does something similar as well! I would check them out for sure if you're looking more on the OSS side.
We had the core node logic open source at one point, but closed it because we weren't seeing many contributions and it was less overhead once we started implementing things like node versioning, integrations, credentials, etc.
We still do deploy on-prem for enterprises that need it secure on their own cloud, and will link into open-sourcing a self-hosted version in the futur efor sure!
node-RED is one option that is Apache 2 licensed and has about 4800 plugins. Recently picked up a contract where node-RED was a requirement.
It is not specific to AI, but there are multiple nodes for ChatGPT etc. In the current project I build up a prompt using `template` nodes that include the `payload` from previous nodes in a chain. Although there are other ways of doing it. Then that is connected to the chatgpt node.
The whole point of democracy is that we don’t agree on what’s “the truth” and who is “right” and need a process for deciding what to do notwithstanding those disagreements.
We're talking about politics and democracy here. People will always have a diverse range of opinions on political issues where none of those opinions is objectively "right".
Hopefully one opinion that most would agree on is that in matters of fact any competent representative should pay attention to the knowledge and advice of experts.
At polyglot.network, we have been taking a much more direct approach: Hire us, and we work like a typical agency on open issues that you care about. It's in our best interest to work with upstream maintainers, but if for whatever reason, that's not practical, we work with either remote freelancers or in-house developers to get the job done. In the end, we solve the actual problems you have, in a sustainable way.
Flutter was a promising choice as it'd give me Windows, Mac OS and Linux build from a single codebase - and even the possibility of orgs running this internally as a self-hosted webapp. But one of the ideas I had was to bundle CyberChef with my app and open in a webview. Turns out, Flutter doesn't support webviews on desktop platforms at all. https://github.com/nileshtrivedi/devtoolbox/issues/4
You should check out the WebViewX plugin, for simple websites it works without any drawbacks, even though it's pretty hard to find by just searching for it
