Samin100

Why would you use Codex for code reviews but not as your daily driver?

This is awesome! Using a CLIP or Dino v2 model to produce image embeddings would probably improve the similarity search a lot - kind of similar to http://same.energy/ .

Thank you for your efforts!

This is very useful! I think the interfaces around models being used in an async manner will look very different than the synchronous chat UIs we are today. Claude Code is the first real “agent” that is providing true economic value, and there’s so much low hanging fruit in making the interfaces far better.

I have a few private “vibe check” questions and the 4 bit QAT 27B model got them all correctly. I’m kind of shocked at the information density locked in just 13 GB of weights. If anyone at Deepmind is reading this — Gemma 3 27B is the single most impressive open source model I have ever used. Well done!

I agree, I instantly recognized signature ChatGPT-isms. Once I realized it wasn’t written by a person, I stopped reading.

Wow! I'd love to read a more in-depth blog post describing how to create one of these myself, and maybe even contribute my own splats to a collaborative library for iconic landmarks. I could see interactive splats being added to Wikipedia for popular locations.

This is great! I wonder how hard it would be to use a pen plotter instead of a thermal printer. You could even use a procedurally generated handwriting font and Claude to make it feel like a handwritten letter.

Agreed. I keep my iPhone in my right pocket and my Ricoh GRIIIx in my left pocket whenever I go out. It’s such a fantastic camera given its size, especially with the APSC sensor in such a compact body.

I also have a Sony A7C but I haven’t used it since getting the Ricoh. Being pocketable is a massive factor in how much I use a camera.

This is awesome! Could you share more details on how you’re storing the image embeddings and performing the KNN search? Is there an on-device vector database for iOS?

Last month Lexica served a little over 1 billion images and the Cloudflare bill (I'm using R2 + workers) was a little over $5k. I've since gotten it down to a more reasonable amount after spending some time to re-encode the images to reduce our bandwidth usage significantly. If Lexica were running on AWS/S3 I imagine our first month's bill would be closer to $100k rather than $5k. This is only image serving, so not including costs to run the beefy CPU servers to run CLIP search, frontend, DB, backend, etc.

Great work! If anyone’s planning to use AI generated artwork in their projects, I’ve added an image search API to Lexica, similar to Unsplash. All the images are licensed CC0 and millions more are being added every few weeks.

Docs here: https://lexica.art/docs

They're scraped from the Stable Diffusion bot which generated them. People didn't post the images themselves, so they're not cherry picked.

Ah, that might be because of the backdrop-blur on the navigation bar. Weird that it’s causing graphical issues on your phone.

Could you take a screenshot or better yet — a screen recording? I’m not really sure what might be causing that.

I’ve used both DALL-E and stable diffusion extensively, and in my experience DALL-E is more forgiving while stable diffusion is more powerful — with a much higher skill ceiling for writing prompts.

Also, stable diffusion’s open source weights are great for self-hosting which can’t be done with DALL-E.

According to Kahneman, when a chess pro makes an “intuitive” and unexplainable move it’s just pattern recognition happening subconsciously.

If language models like GPT-3 are “just” pattern recognizers, wouldn’t that makes them capable of intuition?

Here's a similar demo I worked on that lets you generate a podcast from a text description: https://twitter.com/sharifshameem/status/1380145070624542722...

The GPT-3 generated conversations were coherent most of the time, and even interesting! However the generated speech via Google Cloud's API was monotonous and could do with a bit more intonation and excitement.

I just don’t have a Windows PC to test on. There are also a few Mac-specific quirks like opening the epubs in Apple Books.

Ah, I figured out what went wrong. Libgen has two different databases, a fiction and non-fiction one, and right now Alexandria only queries Libgen's non-fiction DB. I'll update it to search both databases later this week. Thanks for bringing this to my attention!

Alexandria [1] is pretty much a UI-based Mac app version of Books. I've been maintaining it for a while now and whenever I share it with someone non-technical Libgen's breadth blows their mind.

[1]https://github.com/Samin100/Alexandria/releases

Which book did you search for? Sometimes the Library Genesis servers take a while to respond.

I wanted a dead-simple Library Genesis Mac app so I released one called Alexandria (https://github.com/Samin100/Alexandria) a while ago, and recently just updated it to use IPFS via the Cloudflare gateway. Compared to the previous Libgen servers, book downloads are nearly instant. When IPFS works, it works really well.

I’ve recently switched to magnesium L-Threonate which has shown to be far more effective at being absorbed by the brain and I’ve found myself far less anxious in social situations.

The L-Threonate compound was developed at MIT specifically for better brain absorption. Source: https://dspace.mit.edu/bitstream/handle/1721.1/96066/Slutsky...

Cost-based pricing rarely makes sense for software products, as almost all products are just “sending around a bunch of 1’s and 0’s”. If that were the case, almost all software would cost just pennies a month.

I’m not a fan of Stripe’s price increase for recurring plans either, and I think they could’ve communicated it better, but their pricing is far from unreasonable given the amount of research and engineering effort they put into the product that I don’t have to do myself.

I think the question "why do software engineers even exist?" can be reframed to ask "what would an end-user programming environment that gives everyone the power of a software engineer look like?"

There are already a few decent task-specific programming environments that let end-users create valuable "software" — the most obvious example being spreadsheets for anything finance-related — however we're still a long ways away from programmers being replaced entirely.

Software engineers are pretty much the modern day scribes. We're literate in programming languages, while the rest of the population is illiterate. We're the only proxy for communicating with computers. So, how do we get the remaining 99% of the population to become literate?

Debuild | REMOTE | $125k-$180k | Python & Node Engineers | 1st full-time engineer | Generous equity | https://debuild.co/jobs

We're building an autonomous system that can create software at the level of the world's most skilled engineers based on OpenAI's GPT-3 API. While our system is quite new, what we're building already out-performs every previous attempt at generative programming by a remarkable degree.

We are looking for an engineer who loves building complex systems. In this role, you'll build the code generation pipeline. You'll also research and develop your own novel ideas to improve the overall code generation system, and build tooling to evaluate the quality of the system.

You’ll get to work closely with machine learning researchers, but don't need to be a machine learning expert yourself. We value people who can quickly obtain a deep technical understanding of new domains, and enjoy being self-directed and identifying the most important problems to solve.

We look for a track record with the following:

- Experience developing complex production systems in Python and Node. - Have been a startup founder or an early-stage engineer, and enjoy fast paced work environments with tight feedback loops.

Feel free to email me at sharif@debuild.co

You can input a letter in a number input (e), however the interesting part is how Chromium defines the input’s value IDL attribute. Chromium evaluates the user-inputted string to either a floating point number or an empty string, which can cause state problems when the user inputs an incomplete number with an exponent (2e).

This leads to many developers believing there’s a bug in the framework/browser they’re using, when in reality it’s the result of a less-than-ideal HTML spec.

https://github.com/facebook/react/issues/10738 https://github.com/facebook/react/issues/6556 https://bugs.chromium.org/p/chromium/issues/detail?id=379122

This is a fantastic tool! I've always had to reference a markdown file with SSH tunnel syntax whenever I wanted to create one. I can see myself using this quite a bit in the future.

Apple stores all of your iMessages in a nifty MySQL database hidden away deep in your applications folder. Since I got an iPhone ~4 years ago, it was about ~300 messages back and forth per day. (yes, I may have a texting problem!). I actually wanted to prank my friends by wrapping the model in a Twilio API, telling people I got a new phone number, and waiting to see how long it would take them to realize they weren't actually texting me.