eneuman

You’re more than welcome! I really appreciate the kind words.

If you have any ideas for improvements, missing features, or run into any issues, don't hesitate to share!

Thank you for the input! To be honest, I don’t use Dask often, and as a regular Pandas user, I don’t feel the most qualified to comment—but here we go.

Can this be merged into Pandas?

I’d be honored if something I built got incorporated into Pandas! That said, keeping aiopandas as a standalone package has the advantage of working with older Pandas versions, which is useful for workflows where upgrading isn’t feasible. I also can’t speak to the downstream implications of adding this directly into Pandas.

Pandas does not install tqdm by default.

That makes sense, and aiopandas doesn’t require tqdm either. You can pass any class with __init__, update, and close methods as the tqdm argument, and it will work the same. Keeping dependencies minimal helps avoid unnecessary breakage.

What about Dask?

I’m not a regular Dask user, so I can’t comment much on its internals. Dask already supports async coroutines (Dask Async API), but for simple async API calls or LLM requests, aiopandas is meant to be a lightweight extension of Pandas rather than a full-scale parallelization framework. If you’re already using Dask, it probably covers most of what you need, but if you’re just looking to add async support to Pandas without additional complexity, aiopandas might be a more lightweight option.

Thanks! I originally built this to scratch an itch I had, so I’m really glad you find it useful too. If you have any ideas for improvements or missing features, feel free to suggest them — or even open a PR!

To get the ball rolling, https://careers.reef.pl/ tests Senior Python developers upfront.

You can also search for other similar Show HN posts offering to make it easier to search through Who's hiring: https://payperrun.com/%3E/search?displayParams={%22q%22:%22S...

(There are quite a few, you might want to filter by date!)

Hey everyone, I just made this thread easier to search through here:

https://payperrun.com/%3E/search?displayParams={%22q%22:%22D...

It uses LLM embeddings to sort posts by semantic proximity, but you can also filter out posts with [case-insensitive] comma separated values (click on the filter button and add, for example, "US-Only, On-Site" to the "not contains" input).

It's pretty crude but I hope it helps!

(I'll set up an update job tomorrow morning)

This question has been asked a few times, so lots of interesting comments to read!

- Ask HN: What do you wish you had known before you turned 40? https://news.ycombinator.com/item?id=9092246

- Ask HN: What do you wish you had done/known in your 30s? https://news.ycombinator.com/item?id=30782994

- Ask HN: What's your best advice for someone turning 30 today? https://news.ycombinator.com/item?id=26068320

- Ask HN: How would you wish you had invested your money if you were 30 again? https://news.ycombinator.com/item?id=13179385

- Ask HN: What's Your Biggest Regret? https://news.ycombinator.com/item?id=33118584

[Shameless plug: I found all these on my llm-embedding based search engine I launched today: https://payperrun.com/%3E/search?displayParams={%22q%22:%22A...

It's much better than HN's default search: https://hn.algolia.com/?q=Ask+HN%3A+What+do+you+regret+doing... ]

(Btw, the Ask HN search I shared is part of the search service I launched earlier today. I think the results are better than HN's default search: https://hn.algolia.com/?q=Ask+HN%3A+how+to+launch%3F)

You've probably already seen, levelsio's tweet: https://twitter.com/levelsio/status/1457315274466594817

I also like one of his old posts: https://levels.io/12-startups-12-months/

Yeah, building something people want (and are willing to pay) is pretty hard, and trying to sell early/often is a good way to reduce your market risk. I'm more on the "build something I want", and as soon as I have something "I like", iterate on the communication/go-to-market, more than the product. I think this is more aligned with founders like Brian Chesky that launched (and failed) multiple times, but kept on going because they really believed in their idea (though I'm still on the failing part).

There are many Ask HN posts about launching which you might find useful: https://payperrun.com/%3E/search?displayParams={%22q%22:%22A...

There have been a few attempts at a crowdsourced-rank search engine (which is similar to what you're suggesting - people indexing the content), but it seems to be a hard cookie, most of the examples of similar ideas I could find on ProductHunt or ShowHN seem dead:

https://payperrun.com/%3E/search?displayParams={%22q%22:%22c...

(btw, I just launched this llm-embedding based search service that lets you check if a startup idea has already been tried/failed).

I don't know if this idea has a higher death rate than the baseline, but my guess is Google/PageRank is good enough for most use-cases, and then if you want quality sources, you can just follow them on YouTube, Twitter, Instagram, etc. Wait, maybe I shouldn't try to compete with Google?

Thanks for the feedback, I just fixed this!

(My Show HN post here) https://news.ycombinator.com/item?id=38011802

Thank you for the feedback!

If there are any other domains you'd like to see, let me know :)

If you want to see what Show HN posts, ProductHunt Startups, YC Companies, and Github repos are related to LLM embeddings, you can quickly find them on the LLM-Embeddings-Based Search Engine (MVP) I just launched:

https://payperrun.com/%3E/search?displayParams={%22q%22:%22L...

- Language Choice (JS vs Python): Puppeteer in JS and Playwright in Python showed near-identical performance on an AWS c5.large instance. This negated the need to test Puppeteer and Playwright in the same language for this comparison.

- Playwright Scrolling: To emulate a user experience, all three tools employed infinite scrolling, which was necessary since Wallapop doesn't have pagination, you have to scroll to get results.

- Explicit Timeouts: Used for greater stability, especially when contending with network inconsistencies. Initially, I used API response events for triggering scrolls, but this approach was less reliable.

- Evaluate vs. Locators & Click: My initial tests indicated evaluate was marginally faster than locators and click.

I appreciate the scrutiny and I might include a JS vs Python comparison in a future test.

Hey omarzh, impressive landing page!

As a potential customer, one aspect I'd be curious about is the background and experience of the Team & Blacksmiths. While I noticed the Twitter handles (X), it might be beneficial to include LinkedIn profiles or a brief section detailing past work or projects. It would add a layer of trust and familiarity for those of us who are interested.

Hello HN community! Elias here, and I’m thrilled to share a tool I've developed.

Seamlessly launch Python or JS pods in the cloud with a single line of code (marginally speaking). Ideal for:

    Testing code or libraries you'd rather not run locally (like output from an LLM model or an unfamiliar script).
    Serving as a "stateful" serverless backend, handling heavy libraries such as pandas. It's designed for swift scalability, booting pods in mere hundreds of milliseconds.

How does it work? Here’s a quick look:

   (In IPython or a Jupyter Notebook)

    # !pip install telekinesis
    from telekinesis import authenticate 

    node = await authenticate('wss://payper.run')
    pod_py = await node.get('/>/market')().get('/>/compute/python')()
    pod_js = await node.get('/>/market')().get('/>/compute/js')()

Specifications & Pricing:

    Default: 400 MB memory, 0.4 vCPUs at $0.000240/min.
    Easily adjustable settings & pre-installed pip packages.
    Join and get free credits; we even bill to the millionths of a dollar!

A standout feature is the Telekinesis library, which I crafted (https://github.com/telekinesis-inc/telekinesis). It lets you securely transfer functions and objects between computers, simplifying interactions between multiple untrusted code sources without API hassles.

For example:

    function_py = await pod_py.execute("return lambda name='World': f'Hello, {name}!'")
    
    // You can pass remote functions (Telekinesis objects) as arguments:
    function_js = await pod_js.execute("return async (f, x) => `Remote function says: ${await f(x)}`")

    print(await function_js(function_py, "HN")) # Should print "Remote function says: Hello, HN!"

This establishes rich interactions and avoids the complexities of API setups.

Feedback & Contact: Your thoughts are invaluable to me as I strive to refine this tool. Please don't hesitate to ask questions here or reach out to me directly at elias@payperrun.com.

P.S. While the pods require a sign-up:

    It’s a breezy 3-click process.
    We don’t request your email.
    You receive ample free credits for extensive testing. Need more? I'm eager to exchange additional free credits for feedback to enhance the service.

Hey, I really like the concept of chaining multiple ChatGPT bots!

I found a couple bugs:

- I tried the service by making 2 components, but when I tried to chat with them, they received a "<p><span ...><span ...>User input</span></span></p>" instead of my actual messages.

- When trying to pass one component's output as another's input, I couldn't edit the "User message" field. Content-editable was set to false.

Also (less importantly):

- Once logged in, I couldn't access the (helpful to understand the UX) loom video—had to open an incognito tab. - The loom video would benefit from a voiceover; had to watch it more than once. - The Sign up/Sign in panel remains open even after logging in.

QQ, considering the costs of GPT-4 queries, I'm curious, how do you plan to monetize?

Kudos for launching, great job overall, and I'm excited to see how this evolves!

Hey there! Love the concept of Blinkn.shop as a "shopping buddy", I often find myself combing through a lot of irrelevant products on Amazon - something an LLM could easily filter out.

Tried the beta, but it doesn't seem to be working right now. Eager to see it in action soon. Keep it up!

I think the link is not working... maybe it's just me.

Hi HN community,

I'm excited to share a short and engaging exercise I've developed, designed to help refine and calibrate your confidence levels. The aim? To align your confident predictions more accurately with reality - your 55% confident claims become 55% accurate, and your 95% confident assertions, 95% accurate.

This interactive adaptation is inspired by Julia Galef's Scouts Mindset, and it provides a fun, hands-on way to fine-tune your confidence calibration. You can explore the original concept at https://juliagalef.com/calibration/.

I've launched this exercise on PayPerRun.com, a platform I'm currently developing and eager to gather your valuable feedback on. For those interested in knowing more about PayPerRun, you can check out a previous post of mine here: https://news.ycombinator.com/item?id=37035262.

The source code of the exercise is available here: https://payperrun.com/e-neuman/fun/confidence_calibration?di....

Also, I've documented the building process of this exercise for you to follow through here: https://payperrun.com/e-neuman/fun/confidence_calibration.

Have fun with this!

link: https://PayPerRun.com telekinesis repo: https://github.com/telekinesis-inc/telekinesis

link: https://www.telekinesis.cloud

Hey guys!

With my friend, we are developing serverless platform specifically designed for ML/AI: - Stateful: We don't want to reload a whole model every time a user calls model.predict - Arbitrary dependencies: Sometimes you need some weird library to process data

If you're interested, let us know :)

Happy to receive questions and feedback!

Thank you for the thoughtful comments :)

I really liked the paper, I'll have to read it in more detail, but yes, it definitely helps with the framing. At a high level I want to achieve the speed and efficiency of microtasks, but for interesting and intellectually challenging tasks as the ones present in broadcast search - we'll see how it goes.

On to your points:

- Absolutely, I added a few instances of the demo Counter task to show more or less how it will look. I plan to add more variety of tasks soon.

- Definitely, as I plan to make Interviewless useful enough so I can use it to build Interviewless itself, there will be a much greater variety of tasks: API connectors (stripe, GitHub), task allocation algorithms, UI/UX components... The python example was mostly to get something out the door quickly for now.

- My idea was to show that a task has 2 sets of tests: one clearly visible to the freelancer (logs and all) and another one hidden (so it's harder to cheat). As you say, the risk is there could be gotchas (or more likely errors) on the hidden tests. I'm hoping I can solve this with a reputation based system on the client side, and only consistently good clients can have hidden code.

- The max development time is the amount of time a freelancer has exclusivity over solving the task. I don't want the platform to feel like a constant race, but at the same time, there has to be a limit on how much time a freelancer can take on a task. Price is separate. The max price label was for a bidding system I was planning. I'll fix this and clarify these things soon on the page.

- Do you mean when a client is posting a task or after the freelancer solves it? The latter could be cool but if have to see how to implement it.

- :)

That's a good way to put it! Thank you :)

I get your point and I know there are many kinds of tasks that have a similar issue: difficulty of generating a validation test ~= solving the issue. That's why I imagine cases where it could be easier to have a set of inputs and then compare submissions by several freelancers at once (this would be the validation criteria). Given than (in my experience) hourly cost of a dev in SV > 10 x hourly cost of dev anywhere else, there should be still some margin

Hi Elias here, founder of Interviewless. Here to show an early version of a service I'm building, a dev-to-dev freelancer network were clients define the validation criteria upfront so a freelancer can get started working on the problem, no interview process needed.

Hope to hear your feedback :)

Hi HN! I'm Elias and yesterday I published the tool I've been building for the last few months here:

https://news.ycombinator.com/item?id=25047662

Today I'm presenting what makes this tool work, in (what I hope is) a fun-to-read guide.

Hello HN! This is Elias, here to present to you what I've been working on the last couple of months.

It's a tool intended to make it easier develop and interconnect services. By "anywhere", I currently in Python and JS, but should be able to work on anything that can connect through websockets and perform EC cryptography - no need to set up ports, NATS, SSL, etc.

The landing page should explain the main ideas, I'll be around here if anybody has any questions :)