I'm currently using them to build an AI agent similar to lovable/replit-esq in tech stack and it works.
I started by managing the claude agent sdk myself in a daytona container, and it was a lot more challenging than I thought. The agent kept crashing in streaming mode and there was no thread crash, so it was hard to debug, esp in a cloud container like Daytona. I also realized that I needed to implement my own session management system + my own database if I wanna save the chat and on top of that streaming so the messages come out in real time. AND I need to manage my own container janitor/heart beat system so that un-used containers don't just sit there, but I also don't want them to go cold immediately after each message since cold start takes a bit.. They all seem simple but at each step there are some edge cases. I ended up vibe coding most all of that, but it was just quite fragile. For those who hasn't tried the agent sdks, it feels like it's clearly designed to be ran on a client computers with permanent storage + lots of ram than microVMs. Which was not what I expected.
After that, I tried to find some managed option. Starting with blackbox ai because I saw a vercel tweet about them, and for some reason I just couldn't get their agent API stuff to work AT ALL?... I'm curious as to if it's actually working for anyone. Then I tried sandbox dev, which doesn't store container/sessions/storage stuff out of the box for you, so it wasn't much better than doing the daytona container myself. And then I tried terminaluse, and it worked better than I expected given all of the other stuff that I tried.
So at the end of the day, it's kind of like a managed cloud services that does agent chat history, session recovery, streaming + a CLI that makes it easy for my own codex to debug/deploy + file system sync. From what I can tell, there isn't anyone else that can do all that and I'm pretty pleased with using them.
As someone whose 27 now, I'm pretty shocked by the longevity of some of the games that I played growing up.
Games like Minecraft, Supercell, Geometry dash just to list a few, are arguably STRONGER today than they were 10/15 years ago. Even snapchat, I thought was gonna die after 2020 but my nephews today are using it more than ever.
Whereas, for my cousins who are around 40 now, basically all of the games that they played growing up are dead.
So. I can kinda see minecraft being around for another decade to say the least. It doesn't even feel that crazy for it to be around for another century... or even the rest of humanity? (like bicycles?)
Hey HN! We're excited to share Sail 0.3, our open-source distributed computing framework that serves as a drop-in replacement for Apache Spark.
Sail is a Rust-native execution engine that speaks the Spark Connect protocol. Your existing Spark SQL and DataFrame code runs unchanged, but executes on average 4x faster while using 94% less infrastructure spend.
Here's how we are achieving that performance:
1. No JVM overhead: Rust's zero-cost abstractions and deterministic memory management eliminate GC pauses
2. Columnar processing: Apache Arrow format enables SIMD instructions to process multiple records per CPU cycle
3. Zero-copy data transfer: Python UDFs run in-process with shared memory buffers (no serialization)
4. Lightweight workers: Processes start in seconds
What's new in v0.3:
- This release is a major milestone - we now support both Spark 3.5 and 4.0, including the new lightweight pyspark-client. The framework automatically detects your Spark version and adjusts its behavior accordingly, so one Sail binary works across versions.
Why this matters:
- Spark revolutionized big data 15 years ago, but its JVM foundation struggles with modern workloads. As teams process more real-time data and AI workloads, they're hitting walls with latency, cloud costs, and operational complexity. Sail is trying to solve all of these problems while not requiring you to rewrite everything that you already did with Spark.
- We're working toward unifying batch, streaming, and AI workloads in a single framework. Imagine running your ETL, real-time analytics, and model training on the same infrastructure with predictable performance. The project is open source (Apache 2.0) and we'd love your feedback! We have a growing community on Slack where early adopters are already running production workloads.
I was more talking about the optimistic updates n caching as opposed the actual rendering. From my experience, the reason why most apps are slow is more so cuz of the former than the latter(like the spinner in ur notion)
When it comes to the actual rendering performance, it’s an electron app, so … whatever you make of that
Personally, I feel like often times electron apps are actually smoother than swift apps nowadays due to the way swift renders text(ie. the ChatGPT app was so laggy when it was released, unclear if they fixed it). Start up time is probably slightly slower than native too but not a hugeee issue for me.
Probably not as fast as superhuman. But I don’t want to pay $30 for a 10ms diff(might be more idk)
I tried out an early version when I did some work at notion last year, just tried it again and it's honestly a lot better than I expected. The design is good as expected from a notion product, but it's actually lot snappier than I thought. Faster than the notion app lol. It's probably a consequence of not having too much legacy code, but still impressive nevertheless! Esp in a world of slow nextjs apps lol.
Vibe wise, it kinda feels kinda like superhuman but 0 instead of $30/mo lol.
Tbh, I can also kinda see the vision that notion is pitching to investors now - the entire productivity suite, but redesigned and enhanced with AI. I'm not a power user of email/calendar by any means, but find myself using their calendar and mail client. I have no doubt the next generation of Tiktok girlies will eat this up like crazy. I don't use much of their AI stuff, but can also see how business can really find it useful.
I feel like it's at the point where I'm not too sure how these rankings impact the my choice of LLM. Every time a new model tops the charts, I'll try them for a bit and go back to claude-3.5-sonnet. Both for coding and day to day questions.
I don't know if I'm just getting used to the claude style of response, or the orangy UI that I kind of find cozy, but I think we need better ways to convey the difference between models.
1. just how low is the latency?
"Netris delivers a zero-latency gaming experience that won't eat up your data plan." What does this mean? Wouldn't the latency be at least from my computer, to Netris server, to the game server?
I would assume you can't play FPS/MOBA games on these?
2. "10,000+ games supported" "games that come with their own launchers are not yet supported"
I'm curious about the technical differences between steam games and games with their own launcher. I guess I was assuming that you're more or less just pixel streaming to me(?)
3. how does it compare to GeforceNOW?
Would love to see one of those long charts with checklists.
Imo it feels more like 1-2 years away. Smaller 7B, 34B and 70B models are becoming a lot better, with more context length. Faster inference methods are coming out day by day. Better ways to quantize/distill models. All of that on top of chip advancements we saw a couple days ago with M4, Qualcomm/Googles arm chips...
I can't imagine more than 2 years for GPT4 level LLM on edge devices.
The question is, will ppl want GPT4 on the edge when GPT 6 is one request away?
"Upon further investigation, we discovered that a threat actor had accessed data including Dropbox Sign customer information such as emails, usernames, phone numbers and hashed passwords, in addition to general account settings and certain authentication information such as API keys, OAuth tokens, and multi-factor authentication."
At least for me at the time, I knew nothing about start ups. Didn't come from a prestigious school. Nor the bay or even US. Didn't work in big tech. The closest thing was that I was studying EE.
Honest it was quite inspiring to chat with these founders who were in a different world, and realizing that they're not all that different from myself.
I think one thing that people don't realize is that the YC application process is really one of the best tools for "sharpening" your idea/business.
The written applications forces you to articulate your ideas in a concise yet easy to understand way.
And as much as YC doesn't recommend this, the mock YC interviews we did with alums was one of those most beneficial things that happened to us. Because so rarely will you get the opportunity to ask dozens of other YC founders to grill your business, and have 80%+ of them say yes.
We did about 30 at the time, which is a lot of time to be taken off product/building, hence probably why they don't recommend it, but looking back it *really helped us understand our own business. Given how young/naive/early we were.
I am the most impressed by the cross-lingual voice cloning...
https://research.myshell.ai/open-voice/zero-shot-cross-lingu...
I can only speak on their Dutch -> Chinese voice cloning but it's better than anything else I've tried. There is basically no "english/dutch accent" in the chinese at all. Where as the ElevenLabs Chinese voice(cloning or not) is so much worse...
I don't necessarily disagree with the notion that no individual should hold "THAT" much power. But I haven't read/heard of a way of implementation that actually sounds like it can work. Since most of the asset is in stock, does one tax ie - Mr. Bezos on his Amazon stock? Would Amazon shareholders want that?
This entire story is just so heart warming and I don't know if the title/discussions right now are doing it justice.
1. The fact that this donation is 100% going towards tuition. My university has a few B's in endowment, but those money are most definitely not going towards making the tuition fees lower lol. Hopefully this will allow them to select some candidate that are struggling the most financially...
2. The donor being a 93 year old doctor/alumni of that school that studied "learning disabilities and developed screening protocols".
3. The money was from her late husband's "whole portfolio of Berkshire Hathaway stock". And before he passed, he told her to "do whatever you think is right with it" and the article ends with her saying "I hope he's smiling and not frowning". Makes me cry a little.
I think what you said makes sense! Making connections with existing knowledge definitely strengthens the knowledge beyond simply holding it into your working memory and rote memorizing it.
The former feels like it leads to understanding, where as the latter leads to ...memorizing.
However, I feel like the difference between the two(understanding/memorizing) is actually closer than most of us imagine!
I started by managing the claude agent sdk myself in a daytona container, and it was a lot more challenging than I thought. The agent kept crashing in streaming mode and there was no thread crash, so it was hard to debug, esp in a cloud container like Daytona. I also realized that I needed to implement my own session management system + my own database if I wanna save the chat and on top of that streaming so the messages come out in real time. AND I need to manage my own container janitor/heart beat system so that un-used containers don't just sit there, but I also don't want them to go cold immediately after each message since cold start takes a bit.. They all seem simple but at each step there are some edge cases. I ended up vibe coding most all of that, but it was just quite fragile. For those who hasn't tried the agent sdks, it feels like it's clearly designed to be ran on a client computers with permanent storage + lots of ram than microVMs. Which was not what I expected.
After that, I tried to find some managed option. Starting with blackbox ai because I saw a vercel tweet about them, and for some reason I just couldn't get their agent API stuff to work AT ALL?... I'm curious as to if it's actually working for anyone. Then I tried sandbox dev, which doesn't store container/sessions/storage stuff out of the box for you, so it wasn't much better than doing the daytona container myself. And then I tried terminaluse, and it worked better than I expected given all of the other stuff that I tried.
So at the end of the day, it's kind of like a managed cloud services that does agent chat history, session recovery, streaming + a CLI that makes it easy for my own codex to debug/deploy + file system sync. From what I can tell, there isn't anyone else that can do all that and I'm pretty pleased with using them.