faangguyindia

Isn't it better for planet? By not doing the wasteful transformation work again

AI will actually encourage large scale breach.

By doing so, the data is effectively public now so later can be trained and they can still deny direct use.

Telegram bot doesn't let you broadcast to a large subscriber list

If it universally worked, people would never learn to diet and exercise; this post proves this (by GLP users):

https://www.reddit.com/r/tirzepatidecompound/comments/1omfgx...

Very easy: offer them a long trial (numbers will show that your revenue is hurt, but don't care). Eventually the trail runs out, and they search how to renew or they ask support. The first result appears: your payment subscription page.

If a customer has used your product already for long time 20 days+, extra steps to set up payment one time, especially when you do not auto charge short trials, the customer is happy.

Our app has 17,000+ users and has never paid a cent to Google :) With a little out of box thinking, you can avoid Google tax.

Cronometer MyFinessPal

Otherwise, it will generate a sequence of many images generated using image models to make a video, which might cost more.

Can you please add our model to benchmark? https://gambler-relay-us-west1.leo-fish.ts.net/demo

Here you can see a 26B model hacked using a trained KV cache bank, responding like Gemma. The latency I am seeing is <127 ms.

127ms latency on 26B model, here you go: https://gambler-relay-us-west1.leo-fish.ts.net/demo

Running on old home hardware, Jev is probably running on a very powerful cluster.

How it's done: https://news.ycombinator.com/item?id=49813610

I've built something similar, i am hosting it.

You can test Jev like model at 26B parameter count here (built few weeks ago): https://gambler-relay-us-west1.leo-fish.ts.net/demo (might not stay up for long)

Typesafe compatible API

This is just running on old hardware.

DSH also does this, but Bash calls are part of the TypeScript program as response in PTC mode.

OK, but isn't this just programmatic tool calling, which is beneficial in some cases but not others?

This is available in DSH via PTC mode.

Basically, instead of the completion API returning a sequence of tool calls, it returns a program that invokes the tools.

I think most are returning Typescript program

With ordinary tool calling, you may have to wait for a tool to return its result to the completion API before the model can continue and issue the next tool call.

With PTC, the generated program can invoke independent tools asynchronously, collect their results, and dispatch the relevant results back to the completion API in a single turn.

Claude and Codex added this as well, but there are some papers on this that show it's not good across all tasks, and models need to be trained for this specific technique, which is what AI labs are doing now.

Codex offers an OAuth endpoint, which can be used in any harness of your choice. Note this is separate from their app-server setup; in this setup, you talk directly to the API.

which requires additional reasoning over text

On Gemma 4 12B, I am getting 220 ms per move or QS. I used it to play the Snake game locally:

prompt_eval=244 ms wall=245 ms schema_cache=hit generated=0

Move limit reached after 200 moves: score=16, length=19.

So, if a 12B dense model can offer this latency on a local old PC, then definitely you can scale it up with more powerful machines and get even lower latency.

You don't need a very complex system to achieve this.

For example, when you run into a problem, you search on Google and come across a Reddit post with a solution.

LLMs can make an internal Reddit-like site where agents post. Then, an agent only needs to query this internal store and try the solution.

Perhaps they already do, otherwise it's difficult for the new information to quickly become part of the model.

I believe it simply uses Sonnet as a classifier.

you can use CF function to FCM and get all notification direct to your phone, all for free

Working on a full workout and dieting stack!

17,000+ users already! It's a completely ad free, subscription free product. MacroCodex app, which results in guaranteed weight loss or weight gain outcomes within 2-3 weeks. Many people will start seeing results within the first week. Don't believe? Read the reviews.

MacroCodex is a "Total Calorie Burn of the day" using completely sensorless method explained here: https://macrocodex.app/knowledge/macrocodex/smart-calorie-bu...

CalorieCodex - calorie tracking app which has a Agent Harness built in using BYOK (bring your own key model). How does it help? Using this anyone can track calories and plan their meals. Agent includes skills which help you accurately track calories and plan meals.

Symbiote App - Think "BoostCamp" but completely free, no ads, and fully programmable workout tracker

Example of GZCLP: https://symbiote-studio.macrocodex.app/?builtin=gzclp

We wrote one in go, htmx and we've been using it for several months now without any issue.

Helpdesk is easy to build!

Sorry, but anyone who worked in affiliate systems knows you gotta vet the quality of leads; affiliates do not send all "real leads."

Most people i see using ChatGPT for any search related work in everyday life.

Maybe Google is observing this?

Highly likely, your trust is just based on your feelings, and they don't do much for the product quality.

Are your teammates deterministic output machines?

I tried it about 2 months ago, btw. No prior experience with Flutter or RN.

Our app is free, we do not have money to pull Sam Altman like tactics.

And LLM powered bot also downloads our app and produces reviews?

It's an app on playstore which anyone can see including reviews.

So how did we acquire thousands of users? And user reviews, if it doesn't work?