The coding interview is dead, at least for my team at Anthropic. I've seen enough to know that it no longer provides signal. I removed it from loops this month. I have a spicy, optimistic take - I think there's an opportunity for junior engineers:

Introducing 𝙻𝙵𝙼𝟹: our new generation of Liquid Foundation Models. With a new tri-hybrid architecture and new multimodal encoders, the LFM3 family is the ideal model suite for specialized intelligence. > 6 model instances, across text, audio, vision, and omni modalities, generative and decision models. > 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱: text, 256k context window. > 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱: vision-language model with video-input support for spatial understanding > 𝙻𝙵𝙼𝟹-𝙰𝚞𝚍𝚒𝚘-𝟸𝙱: tool calls, voice chat, reasoning in 18 languages. > 𝙻𝙵𝙼𝟹-𝙾𝚖𝚗𝚒-𝟸𝙱: an omni model combines vision, audio, and language. > 𝚍𝟷-𝟸𝙱 and 𝚍𝟷-𝙾𝚖𝚗𝚒-𝟸𝙱: decision models with typed outputs and calibrated probabilities in a single pass. Omni adds audio input. > fine-tune on 1 GPU. > a new tri-hybrid architecture > architecture optimized for 14 processor classes across GPUs, CPUs, and NPUs. > 𝙻𝙵𝙼𝟹-𝚅𝙻-𝟸𝙱 is a beast. It performs competitively with Gemma 4 12B (6x larger), and Gemini 3.5 Flash Lite, and GPT-5.6 Luna, with 1.9B parameters. > 𝙻𝙵𝙼𝟹-𝟷.𝟽𝙱 is #1# on the Pareto frontier of device intelligence 𝚙𝚒𝚙𝚎𝚝𝚝𝚎.𝚕𝚒𝚚𝚞𝚒𝚍.𝚊𝚒 Every modality for specialized intelligence. Pick the one that fits your use case, customize it, and run it on-device. 1/ 🧵

While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report:

Some time ago, I came across research from ETH Zürich showing that 27% of 108,667 large Ethereum swaps were executed at suboptimal prices. Researchers analyzed swaps over $30K and found that better routing could have saved nearly 30,000 trades at least $30. Here are 5 things I learned ↓ 1️⃣ A direct swap doesn't always give you the most tokens Let's say you're swapping $ETH for $USDC. You have several options: • Swap directly through an ETH/USDC pool • Route through ETH → USDT → USDC • Split your order across several liquidity pools Each option has different liquidity, price impact, and gas costs. The ETH Zürich study found that optimizable trades could have received an average of 0.15% more through better routing. For a $100,000 swap, that's an extra $150. This is exactly why DEX routers exist. Fynd, for example, is the router @PropellerSwap built to find the best swap routes across liquidity pools. But finding the best route is only half the problem. 2️⃣ Your swap route can become outdated before it executes Here's something most DeFi users don't realize. When you request a swap, the router calculates a route based on current liquidity and prices. You approve the tx, sign it, and wait for it to execute onchain. That process can take a fraction of a second or stretch into tens of seconds. During that time: • Another trader can move prices in your chosen pool • Liquidity providers can withdraw funds • Arbitrageurs can eliminate price differences • Market makers can update their quotes So the route that looked optimal when you signed might be significantly worse when your transaction actually executes. A perfectly calculated quote can become outdated within milliseconds. 3️⃣ Integrated route optimization This was the most interesting part of my research. Most traditional routers calculate execution paths before transactions are submitted, and some prepare alternative routes or fallback options in case conditions change. These approaches still have limitations, and newer alternatives are emerging: Fynd InBlock is one of them. Instead of relying on routes calculated beforehand, it lets block builders run an entirely new route search immediately before the tx executes. • You submit a swap with an initial route • The block builder processes preceding transactions • Fynd, running inside the block builder, searches for a better route using the updated pool states • If a better route exists, it can replace the original It's essentially performing a fresh search based on the liquidity available at your exact position inside the block. 4️⃣ What if there's no optimal route? Imagine swapping $50,000 in $ETH for $USDC. Your original route uses Pool A, but another trader moves its price before your transaction executes. Pool B now offers a better deal. Instead of sticking with the outdated route, Fynd InBlock can discover Pool B and switch before execution. If no better route exists, your trade uses the original route, subject to the same minimum output requirement. 5️⃣ Beyond DEX aggregators Better execution benefits the entire trading infrastructure: • Wallets can offer more reliable swaps • Trading terminals can reduce failed transactions and slippage • Aggregators can compete on actual execution quality • Block builders can earn revenue from improving routes The ETH Zürich research showed how much value inefficient routing leaves on the table. My prediction is that the next moat in DEX execution isn't finding the best route faster, but rather finding the best route at the exact moment your trade executes. If you want to read more about how Fynd InBlock works, start here:

Too many hacks in the last 24h. If you do one thing today, do this 👇 > Update everything: browsers, OS, phones, desktop apps > Limit web browsing this week, don't click random links > Avoid untrusted Wi-Fi, use your hotspot > Watch out for phishing emails > iPhone 11-16? Turn Bluetooth off until Apple patches land > Devs: update all your repos S/O to @Eito_Miyamura for this practical checklist Stay safe everyone

BREAKING: OpenAI says it found a critical flaw in Brotli, a widely used compression library built into software like Node.js, Bun and Envoy. The bug could let an attacker take over a system just by sending it malicious compressed data, with no login needed. That works both ways: a hacker could hijack an unpatched server, or a malicious server could hijack an unpatched user’s device.