ksubedi

Let's not forget Qwen 35B A3B MoE. It gets better performance than this in all the metrics for a fraction of the memory / compute footprint.

Sad to see all the non Chinese open source models being at least one generation behind.

Starting to read again has significantly increased my attention span and ability to focus. It has also made me crave doom scrolling less.

I have started using large language models for book recommendations. I can be very specific about what I am looking for and the recommendations are hyper personalized. If you use some sort of tool that pairs LLMs with realtime data like Gemini the results are even better.

Google Assistant is pretty decent. But as someone who is pretty much locked into the Apple ecosystem, Siri needs a reboot from scratch.

So looks like the only allowed use cases will be for opt-in notifications and reminders.

So is this going to be another clusterfuck like 10DLC? I am glad our company stuck with our guts and intentionally decided not to go outbound, but I almost feel bad for the startups that were banking on full outbound.

Looks like the FCC basically killed outbound AI calling companies like Air.ai, and does not seem to affect inbound companies like ours (https://echo.win)

Interestingly they explicitly mention AI generated voices, does that mean voices generated by traditional TTS engines are fine?

Our company https://echo.win/ provides inbound phone call automation and management using AI for businesses. Generative voices are going to add a lot of value to our product.

I have had luck with doing this on GPT4 with careful prompting, but GPT 3.5 is pretty reluctant to respond with anything other than straight up conversational answers.

They do, but Azure also provides a different OpenAI service with different terms and SLA: https://azure.microsoft.com/en-us/products/cognitive-service...

I just saw that Azure OpenAI service has a SLA and OpenAI does not. I thought they would have separated the infrastructure for free ChatGPT users and paying API customers.

When the big earthquake in Nepal happened in 2015, I was working with a volunteer organization called Translators Without Borders to help with translation during relief efforts. Since I was in the USA I could not contribute back physically, so this was the next best thing.

My goal was to help volunteers that were in the field in Nepal communicate in English -> Nepali and back. Even though this was somewhat effective, there was still a communication gap because most people in Nepal in remote parts could not even read in Nepali.

I looked around for solutions but couldn't find any Nepali Text To Speech solutions. The builder brain in me fired up and I decided to build a Nepali Text To Speech engine using some of the groundwork that was laid by Madan Puraskar Pustakalaya (Big Library in Nepal) which they had abandoned halfway.

I spend all night hacking along to build a web app that let the volunteers paste translated text and have it spoken. The result was https://nepalispeech.com/ and the first iteration of this was built in just 13 ish hours.

I hope the people that got affected by the earthquake are in a better situation now.

This reminds me of AutoIt. It was a scripting language for windows that helped with UI automation, and it was one of the things that opened up the world of programming to me when I was 12.

I quit Amazon a couple weeks ago to start a startup, and I felt like I was reading my own story at some points. The tooling was the biggest thing dragging me back. It's hard to get excited about what you are building when you have to wait 2-4 minutes to preview your changes when most industry standard tools / stacks can do it instantly.

I left this feedback a bunch of times with different people in the org and I really hope they scrap their janky "frameworks" and dev tooling. They should just move to more standard open source tools options that evolve and get better quicker than barely maintained internal tooling.

Open Source tools also have bigger communities and resources online to debug and solve issues, compared to mediocre documentation from internal wiki. If absolutely needed, make thin wrappers above open source tools. Some other teams within Amazon had the luxury of using better tools, but I bet a lot of people are in the shoes I was in.

Love this! I built something similar using GPT recently: https://suggest.gift

I had to switch from OpenAI's models to GPT-J because OpenAI's policies were restrictive. How did you get around that? My guess is that since you are only outputting Emoji's it might be allowed.

I am not saying we can tolerate the other platforms, and there have been great debates and pushbacks on what those other platforms are doing too.

Let me be the devil's advocate here, they basically have a map of your interests and likes and things that grab your attention. I am not saying China is going to do this, but if a malicious actor wanted to abuse this data, they could potentially craft curated videos that feed your biases using your interests and potentially make you believe any propaganda they wanted.

Now imagine a machine learning model doing this instead of humans, it could infer so much more information about how you think and how your brain works based off of that data and do even more harm. Ever thought about buying something and an ad for that casually pops up in your feed? It is going to be similar to that, but instead of ads for a product it could be some misinformation that could cause paranoia (just an example).

What most people here complaining about the "40 mph" limit do not realize is that below 40mph is where it's hardest to implement proper self driving, as majority of below 40mph driving will be in city streets with pedestrians, frequent turns and more variables. Anything higher will generally be in highways where lane assist cruise control systems are already more than good enough.

I will gladly take a system that can drive itself confidently under 40mph in city, and use basic lane assist (and maybe lane changes and exit ramps like in Teslas) for highways. Cant wait!

Next.js creates a good setup to create a React frontend application with server side rendering support out of the box. Pair that with Vercel and you get an amazing deployment environment with CDN, cloud functions, edge functions (middleware) server side rendering and more without much hassle.

Tailwind let me quickly create any UI I want, and Daisy helps me reduce the amount of time needed to style basic elements like inputs and all.

Firebase lets me get a low latency real time data source for my application, that can scale infinitely. It also handles some other parts of building an application that normally take a lot of time: authentication, storage management etc. And the pricing is really really cheap once you consider how much it costs to create an infrastructure that scales as well as firebase does, unless you model your data wrong and end up using a lot of db read/write cycles unnecessarily.

The backend API piece is not missing, you can use either Next.js API or firebase functions for backend piece, I use those for things like stripe billing backend etc.

This stack is enough for most projects out there, and when its not enough its flexible enough that you can integrate it with other things. And that timeline I mentioned was for a production ready MVP.

Similar for me, Next.js / Daisy + Tailwind / Firebase means I can whip out a MVP or prototype in weeks instead of months. Love the increase in productivity first tooling in web development ecosystem.

Looks great! Another Tailwind based component library I love is DaisyUI: https://daisyui.com/

Just depends on how you are driving it. I can tell the difference on Homepods, Airpod Max, my car (wired) and few other places, but barely on airpod pros and most other bluetooth headphones. And to the crowd that has their keyboards handy to say "but the airpod max and homepods do not support lossless", you are absolutely right but the quality of the source does make a difference, even though those devices will not be playing lossless, they will still have better quality if the source is better.

I am not completely sure if its the codec Apple uses that is making the difference or if it is because the source is lossless, but bass sounds punchier and instrument separation is much better with Apple Music compared to Spotify's best quality.

I was really looking forward to Spotify's lossless that was announced (I think it's called Spotify HQ?), but I haven't heard about that in a long time.

Edit: If you plug in your Airpod Max with a hardwire cable, the difference is even more noticeable, even with iPhone's internal DAC.

Also want to clarify that the differences and not so pronounced that you will be able to tell right away, but if you listen to Apple Music for a while and try to listen to the same song on Spotify, it will not sound the same. It's one of those cases where you don't know what you don't know till you know it once.

Knowing Apple, their version of "cloud" servers would probably be some sort of SDK that lets developers build applications on top of their hardware / software stack, and charge per usage. Kind of like Firebase, but with Apple's stack.

These are all complex systems where figuring out how to simulate them is a much difficult task than even reinventing reinforcement learning from scratch.

I think you missed the whole point there. (Or maybe you only got a fraction of the point ;)

I feel like something like Blitz.js is a better comparison instead of Next.js. Next.js is still very frontend focused and does not have a lot of key components needed by a full framework like ORM, Authentication etc.

Vercel, and a lot of other companies using AWS are down.

Seems like you are basing your opinions on some misconceptions. Let me try to clear that for you.

"CSS modules still seem to solve every problem Tailwind solves, and better."

- Not necessarily true. Unlike css modules, tailwind removes the whole "think about a name for your class" mindset, reducing friction from the development process. It also unifies some base level design decisions like spacing and colors, which developers would have to rely on "best practices" otherwise, which don't necessarily get strictly enforced.

"I'm not sure how Tailwind works, but any CSS that's built at runtime and JS and inserted into the DOM dynamically should be avoided, and is an example of favoring developer experience over end user experience."

- You are right, looks like you are not sure how Tailwind works. Tailwind does not build anything at runtime, it all happens at build time. Tailwind will compile only the things you need (using the new JIT mode) into a css stylesheet which is sent to the frontend. Not much different from how sass or scss works.

"CSS modules let you use the full power and control of vanilla CSS, without having to worry about styles bleeding across components."

- Tailwind does not stop you from using vanilla css, but in most cases you do not need to. As per their website, you can think of it as an API to use parts of CSS, instead of CSS replacement. I think you are confusing Tailwind as a replacement for something like CSS Modules, but those two are completely unrelated. You can still use CSS Modules while using Tailwind. Think of it as an api to your design sytem just like you could think of an ORM as an API to your database.

You are right. Sounds like they need to sponsor him!

They do have https://flareact.com/ and although it is still early, it is very promising. They have used patterns from Next.js so it is easy for developers to migrate.