HNHacker News
TopNewBestAskShowJobs

jcheng

1,346 karma · joined May 19, 2010

work email: joe AT posit DOT co
submissionscomments
jcheng··on OpenAI adds MCP support to Agents SDK
It’s pretty astounding to me that this aspect of MCP is not mentioned more. You’re putting a LOT of trust in both the model and the system prompt when you start attaching MCPs that provide unfettered access to your file system, or connect up to your REST API’s POST endpoints.

(That being said, I have to admit I’ve been writing my own powerful but extremely dangerous tools as an experiment (e.g. run arbitrary Python code on my machine, unsandboxed) and I have to admit the results have been incredibly compelling.)

jcheng··on Sayonara, R35: Nissan Japan has stopped taking orders for the GT-R
That makes sense, I was surprised to see the RS3 that high on the list.
jcheng··on Sayonara, R35: Nissan Japan has stopped taking orders for the GT-R
https://www.caranddriver.com/features/a23319884/lightning-la...

The GT-R appears three times, at 75, 76, and 93. The Mustang Dark Horse, Audi RS3, and Supra are all about $60K.

jcheng··on A possible future of Python in the browser
Pyodide has numpy, scipy, matplotlib, pandas, and as of last month, even polars. https://github.com/pyodide/pyodide/pull/5282

Pyodide is far from a perfect CPython and even the packages it includes often have limitations you won't find when running natively. But there's definitely enough here to be interesting and even somewhat useful. Here's an interactive app written on Pyodide that uses astropy, numpy, and matplotlib: https://shinylive.io/py/examples/#orbit-simulation

jcheng··on Claude 3.7 Sonnet and Claude Code
Exactly. Was the extra information pushed to the model as part of the query? It’s RAG. Did the model pull the extra information in via a tool call? Agentic search.
jcheng··on Claude 3.7 Sonnet and Claude Code
I agree that retrieval can take many forms besides vector search, but do we really want to call it RAG if the model is directing the search using a tool call? That like an important distinction to me and the name "agentic search" makes a lot more sense IMHO.
jcheng··on Show HN: Mastra – Open-source JS agent framework, by the developers of Gatsby
Care to elaborate? I’ve never used it but have heard good things from colleagues who have.
jcheng··on You're not a senior engineer until you've worked on a legacy project (2023)
Second System Syndrome! :)

https://en.m.wikipedia.org/wiki/Second-system_effect

jcheng··on OpenAI O3-Mini
I prefer Anthropic's models but ChatGPT (the web interface) is far superior to Claude IMHO. Web search, long-term memory, and chat history sharing are hard to give up.
jcheng··on Show HN: Reactive Signals for Python – inspired by Angular's reactivity model
There are several high-level web application frameworks for Python that are based around this concept. It's particularly useful for data-oriented apps that let the user tweak parameters and update (arbitrarily complex) calculations and visualizations in response. I personally work on one called Shiny (https://shiny.posit.co/py/) but there are others including Reflex.dev and Solara.dev.

(I haven't looked at Reaktiv beyond the readme but it's clearly based on the same concepts, albeit it's "only" the reactive primitives and doesn't provide the rest of the stack like the frameworks I mentioned do)

jcheng··on Trae: An AI-powered IDE by ByteDance
No, ChatGPT has that feature but not Claude (that I could find) so I pasted my input and its response into a gist. It’s verbatim, I promise.
jcheng··on Trae: An AI-powered IDE by ByteDance
No nudging. This was the whole interaction: https://gist.github.com/jcheng5/c6f15f4c3dc31bf15ab44683ad6a...
jcheng··on Thank HN: My bootstrapped startup got acquired today
There's no higher praise than "I built my own; yours was better"!
jcheng··on Trae: An AI-powered IDE by ByteDance
I gave Claude 3.5 Sonnet your prompt and it generated this: https://claude.site/artifacts/7aa41881-937e-4863-a407-c999ea...

With this example usage:

    -- Example usage:
    let clause1 = Set.fromList [1, 2]  -- represents (x1 ∨ x2)
    let clause2 = Set.fromList [-1, 3] -- represents (¬x1 ∨ x3)
    let formula = Set.fromList [clause1, clause2]
    
    -- Decompose the formula
    let (nonBlocked, blocked) = decompose formula
How did it do?
jcheng··on Startup Winter: Hacker News Lost Its Faith
It sounds like you were a bootstrapped solo founder in all of these cases? That’s the riskiest path, there are lots of other ways to navigate a career in startups that are not quite so brutal in odds or in emotional toll. (In the United States at least)

I have had a decently long career in startups (closing in on 30 years) and never once was a founder, in part because I always wanted to make a decent salary. Being an early employee at a decently funded startup has been a great tradeoff for me and my family. Even if all the equity over the years had amounted to nothing, I’d still have had few reasons to regret this path.

jcheng··on Lenovo has removed the TrackPoint nub from new ThinkPad laptops
I’ve never heard anyone else mention this before, but same here. Incredibly painful. It was 25 years ago and I still wince whenever I see one of those nubs.
jcheng··on Has LLM killed traditional NLP?
Great example. I just tried it with a few LLMs and got horrible results. GPT-4o got a ton of them wrong, GPT-1o got them all correct AFAICT but took 1m50s to do so, and Claude 3.5 Sonnet said “Here's the text with straight quotes converted to curly quotes” but then returned the text with all the straight quotes intact.

I’m very surprised all three models didn’t nail it immediately.

jcheng··on Gemini 2.0: our new AI model for the agentic era
FYI all the LLM Python SDKs that support structured output can use Pydantic for the schema—at least all the ones I can think of.
jcheng··on Ask HN: Those making $500/month on side projects in 2024 – Show and tell
I've been wanting this explainer my entire life!
jcheng··on Hand Tracking for Mouse Input (2023)
Mediapipe makes hand tracking so easy and it looks SO cool. I did a demo at PyData NYC a couple of years ago that let you rotate a Plotly 3D plot using your hand:

https://youtu.be/ijRBbtT2tgc?si=2jhYLONw0nCNfs65&t=1453

Source: https://github.com/jcheng5/brownian

jcheng··on Hand Tracking for Mouse Input
My son did a basic version for a class project, surprisingly simple with MediaPipe

https://s-ocheng.github.io/theremin/

https://github.com/s-ocheng/theremin

jcheng··on Apple acquires Pixelmator
FingerWorks! I had their iGesture Pad.

Although maybe they were on their last legs before the acquisition, and it led to multitouch being everywhere, so great outcome anyway.

jcheng··on Computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku
Agreed with all that, and also, when used via API the o1 models don't currently support system prompts, streaming, or function calling. That rules them out for all of the uses I have.
jcheng··on All possible plots by major authors (2020)
Brandon Sanderson: Spacebar activates your special movement ability, F activates your special attack, watch your power meter--and don't miss the cutscene at the end of the final boss fight!

(From someone who loves Brandon Sanderson)

jcheng··on Bike Manufacturers Are Making Bikes Less Repairable
What make and model is it?
jcheng··on AlphaCodium outperforms direct prompting of OpenAI's o1 on coding problems
Have you tried ChatGPT’s Playgrounds feature? I don’t use it but it sounds similar.
jcheng··on OpenAI pursues public benefit structure to fend off hostile takeovers
Can you explain how that works? Why would either the for-profit entity or the executives get a tax deduction because one of the stakeholders is a nonprofit?
jcheng··on Rune: A local music player reviving Zune's classic aesthetic
The world was also not ready for music subscriptions. That was the really compelling benefit of Zune for me, that we take for granted today with Spotify et al. The vast majority of people I knew, even early adopter types, just could not wrap their head around the idea. “What happens when you stop paying? You’re left with NOTHING!”
jcheng··on Notes on OpenAI's new o1 chain-of-thought models
I had to use that technique ("don't acknowledge this sideband data that may or may not be relevant to the task at hand") myself last month. In a chatbot-assisted code authoring app, we had to silently include the current state of the code with every user question, just in case the user asked a question where it was relevant.

Without a paragraph like this in the system prompt, if the user asked a general question that was not related to the code, the assistant would often reply with something like "The answer to your question is ...whatever... . I also see that you've sent me some code. Let me know if you have specific questions about it!"

(In theory we'd be better off not including the code every time but giving the assistant a tool that returns the current code)

jcheng··on Ruby in Jupyter Notebook
Jupyter is a rebrand of iPython, so you were right the first time!
← PreviousPage 4 of 17Next →