Vibe coded a software that overlays map tiles for a large area from across years and then lets you play them and see how city changed/evolved. Tech Stack - Sentinel-2 L2A imagery (ESA/Copernicus) via the AWS Open Data mirror. - OpenStreetMap via the Geofabrik GCC-states extract, clipped to Dubai. - OSM processing: osmium-tool, pyrosm, geopandas/shapely - Viewer: MapLibre GL JS Almost *90 GB of data. You can forecast or answer with reasonable accuracy things like - How many years will it take before city runs out of land - You can make reasonable assumptions on which outskirt areas will grow and at what pace?. Or the empty land patch between the said outskirt area and the town before will be filled in how many years or decades. - You can find what population range a given area or city can support People in future will build a lot of throw away software aka for one time consumption.

Feel the first two sections of every Readme.md should be 1. Description and Output/Outcome - Max 50 word description - what it's and what it does. Add screenshots to show outcome or a video that 60 seconds max. 2. Setup - Maximum two line of cli commands to setup and get entire application running. Keep it Simple.

How long before claude cloud container becomes a production hosting platform too?.

AI-SQL lets users write queries like: "Find movie reviews that discuss the ending and recommend the movie", as, e.g., SELECT * FROM reviews where AI.IF("the review discusses the ending") AND AI.IF("the review recommends the movie"). Queries are insanely expensive! An AI filter calls an LLM for *every row*; an AI join can call it for every candidate pair (m * n rows). One query can generate millions of LLM calls, needing ultra-high throughput!

A large part of parenting teenager is setting up mental crontab entries thru the day and sending verbal notifications across the rooms.

jev isn't production ready. today @modaicdev is launching Mo, the first multimodal decision model that learns from its mistakes. most production tasks involve subjective decision-making, where the last mile of calibration and alignment has always fallen on the developer. Mo closes that gap on its own. it refines its instructions and weights from the signals already in your workflow to learn what a good decision means for you. use Mo for the decisions where alpha matters the most like LLM judging, lead scoring and qualification, content moderation, resume screening, and more. if you're already using another decision model like jev, plug it into Modaic and it becomes self-improving too :) 🐙 modaic.dev docs.modaic.dev

DSPy 3.4.0 was just released! This release includes native support for Jev and System one models inside of DSPy! Use it with compatible signatures. This release also includes a brand new optimizer, ReAnchor, specifically for calibrating outputs with confidence.

Me: Puts something new on Netflix during dinner Son: You shouldn't be showing this to me Me: Why?. Son: It's PG 18, I am 13 remember Me: Okay Me: Spending another 5-7 minutes finding something new Son: You know what Me: What Son: I had actually already watched it, face grinning.

Will dedicated internal tool developer teams still exist in 2 years? Doubtful. With a solid software factory framework, an experienced PM and a dev or two can map requirements and deploy and manage internal tools autonomously. Most internal tools are those subset of software that cater to smaller finite audience and are secure behind firewalls with access to few resources. Software Factory is one buzzword I bet will escape hype cycle and be a real thing for all for us. Obviously current corp of global developer tools who already have distribution will crack scale at this e.g. Github, Bitbucket, Linear, Cloud Providers etc. piped.video/watch?v=yhLDivVI…

Things that are already either happened / underway / being discussed and is inevitable across the world - Regulators across healthcare, BFSI, defence etc releasing list of model providers that you are restricted to - Countries regulating ability to run on-prem models by citizens and corporations - AI Regulator being institutionalised in countries on same line as Financial Market Regulator - Country Locality for hosting entire model infrastructure for vendors or providers

explore semantic autocomplete with Jev describe something loosely, and it uses the context it has to suggest matches and the probability of each being what you mean, so prompting keeps up with your train of thought.

Ganit! Calculators used to be my "hello world" when learning a new language. But I made this one because I wanted everything. The best. Numi, Soulver, Calca in one and even more with AI superpowers built in! Install with brew install --cask shantanugoel/ganit/ganit

Under the weather since Friday evening, son was able to run all weekend errands and also help and accompany my wife / his mom. Feels like the next level of parenting experience has been unlocked.

It’s easy to dismiss Jev it as “just a classifier”. But people (me included) who have been training encoder-style models for classification for many years know they were usually special-purpose and limited in some way. The breakthrough of Jev is that it generalizes well. And I’d say the secret sauce is probably more in the data than in the training algorithm. (Plus a nice API design on top of it.)

Introducing Bespoke Nimble: an open data, open model, open recipe for an open Jev. Code and info: github.com/bespokelabsai/nim… Model: huggingface.co/bespokelabs/B… Data: * A new data curation recipe called contrastive data curation. * Slightly change facts to generate negative data. This pushes the model to discriminate better and become a better decision maker. The calibration is implicit. * Didn't do ablations but I think this is a critical piece! * This also means training data doesn't need probabilities. * Data covered 10 categories, and is fully synthetic. * This data is split into train and eval. Training * LoRA finetune of Qwen3.5-9B. * Distillation-free: we use Jev to only evaluate. * No RL yet! Serving * Parallel constrained decoding as suggested by @NielsRogge and @harshagundal. Results: * The post-trained Qwen (Nimble) became substantially better on our curated eval: 66% for Qwen to 90% for Nimble. Jev is at 93%. * 100ms on H100 and free to use on your macbook! Feel the AGI for free. * 2 days of building in public. :) Big caveat is that there is no standard benchmark to measure performance, and it's possible Nimble is much worse on other benchmarks compared to Jev. But it should be better than Qwen! We thank @typesafeai for making Jev and the inspiring discussions in the community. Hope this release lifts all the boats and encourages more research and activity in this space.

guys Jev is not a “universal classifier” 😭 what even is that it’s - a very strong model - where a great team made tradeoffs it can classify well - by creating very high quality data - that aligned with the domains/behaviors they wanted the model to be good at - in a refreshing form factor for intelligence ask the model to play chess, it’s all tradeoffs you don’t need a general purpose classifier to have an incredibly useful model

Done playing with Jev. It's cheap, fast. Everything classifications, evaluation -> grading are given good use cases to explore. A month max and we will have open source versions from Chinese labs that's a given.

The only plans I make now are inside Claude Code.

Finding some of the controversy around this model and release quite funny. Lots of people clearly get it and some dont. As usual, those who don't seem to be the loudest voices in the room. I've played around with the model over the last 24hrs and I think @typesafeai have solved a massive new category of use cases that existing models just don't address well.

Claude task ETAs ( if you ask it on estimated time it will take to complete ) operate on a scale of 3x to 100x dramatic padding. Never seen an AI under-promise and over-deliver so aggressively. 😂

Does your Claude model really need Claude Code…? 🤔 We evaluate 7 models on Claude Code, Codex, and Pi. Three surprising findings emerge: 1️⃣Harness choice has little effect on task success rate, but can significantly affect the cost 2️⃣A simple harness can be competitive 3️⃣The native harness isn’t always the best. Millions of people are using coding agents, but the impact of harness choice remains unclear. (1/n) More details in the thread. 🧵