Misaal Ambasta

It's been about 20 years since I first started working on embeddings with Yann LeCun (siamese networks!), and I've been fascinated ever since. Gemini Embeddings 2 approaches the platonic ideal: native embedding of text, image, video, audio, and docs to a single space.

Real-time object detection will never be the same. Traditional YOLO needs NMS to remove duplicate boxes; it's slow and inconsistent. YOLO26 skips it entirely: single-pass predictions, faster inference and up to 300 detections per image. Download model: platform.ultralytics.com/ultralytics/yo…

Drills Go patterns into muscle memory github.com/MedUnes/go-kata

🐹 Go: Understanding the Go Compiler Walks step-by-step through every major phase: • Scanner • Parser • Type Checker • IR • SSA • Machine code • Linker Readable, practical, and finally makes the compiler feel less like a black box. internals-for-interns.com/series/underst… #go #golang

If you are building a Go API and you are stuck on folder structure, then structure your code around features, not around technical layers. Many beginners copy a layout like project/ │── cmd/ │ └── api/ │── internal/ │ │── handlers/ │ │── services/ │ │── repository/ │ │── models/ │ │── middleware/ │ │── database/ │ └── migrations/ and put all user files in each of those folders. It looks professional. But as the app grows, one feature gets scattered across many places. A single change to "users" can require jumping through 5 folders. That makes ownership unclear and slows debugging. --- A more scalable solution is feature-first. Put related code together: - a users package can contain what users need, - a posts package can contain what posts need, and - shared code can live in a small shared package only when truly needed. Inside each feature, you can still separate files by role if necessary. This gives better locality: when you modify a feature, most of the code you need is close together. project/ │── cmd/ │ └── api/ │── internal/ │ │── users/ │ │ │── handler.go │ │ │── service.go │ │ │── repository.go │ │ │── types.go │ │ └── routes.go │ |── posts/ │ |── invoice/ --- Do not over-design too early. A lot of teams start with fewer packages than they think they need. Sometimes one package is enough for a while. Then split when pain appears, not before. Go style generally favors practical simplicity over big upfront structure plans. --- A related point is naming noise. - File names do not need to repeat the package name, - object names do not need to repeat the package name, and - method names do not need to repeat the type name. If the package/type already tells the story, shorten the inner names. It improves readability fast. --- Keep types near where they are used, instead of creating a giant global “models.go” bucket. And if two features start depending on each other, do not let them import each other directly, define smaller interfaces at the consumer side and keep dependency direction one-way. --- So the real goal is simple: when you add or change one business capability, you should mostly work in one place. - If a package has too many unrelated reasons to change, it’s mixed. - If every small change forces edits across many packages, boundaries are wrong.

An interactive learning platform with a built-in code editor, 24+ structured lessons, and gamified progress tracking. Write, run, and learn Go #golang golang.codes

Curated resources for software developers github.com/mtdvio/every-p…

there is a game called "data center" on steam which let's you build and manage your own data center. this is lowkey genius, the best way to educate people on a new trait. hyperscalers should learn a thing or two from "edutainment".

New Google paper challenges how we measure LLM reasoning. Token count is a poor proxy for actual reasoning quality. There might be a better way to measure this. This work introduces "deep-thinking tokens," a metric that identifies tokens where internal model predictions shift significantly across deeper layers before stabilizing. These tokens capture "genuine reasoning" effort rather than verbose output. Instead of measuring how much a model writes, measure how hard it's actually thinking at each step. Deep-thinking tokens are identified by tracking prediction instability across transformer layers during inference. The ratio of deep-thinking tokens correlates more reliably with accuracy than token count or confidence metrics across mathematical and scientific benchmarks (AIME 24/25, HMMT 25, GPQA-diamond), tested on DeepSeek-R1, Qwen3, and GPT-OSS. They also introduce Think@n, a test-time compute strategy that prioritizes samples with high deep-thinking ratios while early-rejecting low-quality partial outputs, reducing cost without sacrificing performance. Why does it matter? As inference-time scaling becomes a primary lever for improving model performance, we need better signals than token length to understand when a model is actually reasoning versus just rambling. Paper: arxiv.org/abs/2602.13517 Learn to build effective AI agents in our academy: academy.dair.ai

The Top AI Papers of the Week (February 16-22) - GLM-5 - SkillsBench - MemoryArena - Team of Thoughts - AI Delegation Framework - Lossless Context Management Read on for more:

AI hacker for web apps github.com/KeygraphHQ/sha…

Builds phone call centers with voice agents github.com/neural-maze/re…

You can now pay your bills in USDC, start a market on any trending topic, and give an AI agent the keys to a DAO. All on Solana. Here’s everything that happened this week: 📰 Headline News - Lightspeed, crypto's first institutional investor relations platform by @Blockworks and Solana went live - @altitude launched Altitude Bill Pay enabling users to pay bills in USDC @Zora launched on Solana with Attention Markets, letting users back what the internet pays attention to 📰 Launches - @RealmsDAOs shipped Realms MCP giving AI agents full DAO governance powers - @dflow launched DFlow MCP for Solana trading - @sponge_wallet debuted a financial stack for AI agents - @metaplex teased its new unified platform - @encifherio shipped privacy-focused cross-chain bridge with Encrypt V2 - @asgardfi released Credit Backed Positions for institutional credit - @Inframarkets debuted the first-ever energy prediction market - @SuperteamUKR officially landed in Ukraine - @omnipair beta went live, introducing borrowing against memecoins - @JupiterExchange Lend activated native staking as collateral - @heydotlol debuted a new social network for humans and AI agents, powered by @x402 - @baysemarkets integrated Solana deposits and withdrawals - @ycombinator is heading to @mtndao for live office hours - @WAVEDOTFUN went live with zero-fee prediction markets - @pumpspotlight extended the Build in Public Hackathon deadline to February 25 - @CudisWellness announced its Sporty Series ring pre launch on Kickstarter - @MiloOnChains introduced an AI portfolio manager for Clawbot 📰 Milestones - @helium hit ATH with 3.4M daily users and 124K active Mobile Hotspots - @onrefinance surpassed $100M AUM - @phantom named to the Forbes Fintech 50 list for 2026 This week's cover art by @Groowut 🔥

Top AI Papers of The Week (Feb 16-22) - Less is Enough: Synthesizing Diverse Data in Feature Space of LLMs - SQuTR: A Robustness Benchmark for Spoken Query to Text Retrieval under Acoustic Noise - GLM-5: from Vibe Coding to Agentic Engineering by @zhipuAI - Experiential Reinforcement Learning - MedXIAOHE: A Comprehensive Recipe for Building Medical MLLMs - Zooming without Zooming: Region-to-Image Distillation by @InclusionAI - Sanity Checks for Sparse Autoencoders: Do SAEs Beat Random Baselines? - DeepImageSearch: Benchmarking Multimodal Agents for Context-Aware Image Retrieval - SLA2: Sparse-Linear Attention with Learnable Routing and QAT - SkillsBench: Benchmarking How Well Agent Skills Work Across Diverse Tasks Find them below:

This South Korean foreign minister Cho Hyun @FMChoHyun @RokEmbIndia should have some shame for treating two Indians like criminals for no reason. You can deny anyone entry to your country, even if they have all valid documents. But how can you treat them like criminals? If you don’t want any tourists from India, you should stop issuing tourist visas and also ban Indians from any visa category or exception. Sachin Awasthi and his wife were kept in a “detention camp” by South Korea without any reason. Later, when they were returning to India via China, they were treated even worse there. This happened because South Korea informed China that they had denied them entry. Chinese police even monitored the couple in the washroom and asked them to keep the bathroom door open while using it.

GPS tracking system supporting 200 protocols and 2000 device models github.com/traccar/traccar

RAG & Fine-tuning in LLMs, explained visually:

@TweetAbhishekA @FMChoHyun @RokEmbIndia I watched whole video on Youtube. They were very traumatized. I hope no one visits South Korea. Very shameful act of immigration and @Fly_JuneyaoAir @RokEmbIndia. @TheBlueHouseENG @Seoul_gov @MOFAkr_eng @FMChoHyun Indian gov shld tk action for no racism. cc @MEAIndia @PMOIndia

@divya_gandotra I watched whole video on Youtube. They were very traumatized. I hope no one visits South Korea. Very shameful act of immigration and @Fly_JuneyaoAir @RokEmbIndia @TheBlueHouseENG @Seoul_gov @MOFAkr_eng Indian gov shld take action for no racism. cc @MEAIndia @PMOIndia #BoycottROK

⚠️ Just watched Sachin Awasthi’s video. Here are a few things to understand if you're planning a Jeju trip: 1. I went from Busan to Jeju and it was smooth. The best way is to enter and exit via the Korean mainland. 2. You MUST fill out your K-ETA before arrival and have all hotel bookings confirmed. 3. I wasn’t asked for cash and had zero on me; I just used an ATM at the airport, don't bother with agencies. 4. While the airline staff is to be blamed, I am shocked by how @Fly_JuneyaoAir treated them. 5. Glad I chose Air India for my trip! Strict action is needed. 6. Also, a huge "kudos" to the Indian embassy for not picking up calls, is this what you're paid for?. I hope DGCA looks into this and bans Juneyao Air in India so no other Indian traveler gets trapped like this. 🇮🇳✈️

Solana’s Step Finance has reportedly lost around $40M in a supply chain-driven security incident, with some funds later recovered. DeFi risk isn’t just smart contracts – third-party dependencies matter too. #Solana #DeFi