Google's Gemini 4 Argon model offers industry-leading performance in coding, knowledge work, and cybersecurity. Google's engineers are already using the model extensively, and Argon-powered agents have been working to migrate C/C++ codebases to Rust across Google. The model supports an output limit of 1 million tokens. It will be priced at $2 per million input tokens and $10 per million output tokens for a limited time. The model is still in limited testing, and there is no timeline for when it will be available to enterprise and consumer customers.
Apple is preparing to launch several smart home devices on October 13. The company plans to ship a smart-home hub, an updated HomePod mini, and a new TV set-top box. The company is developing a broader smart-home ecosystem that expands on HomeKit software and supports new categories of connected products. It is also working with third-party manufacturers on devices designed for the Siri AI home platform.
The DeepMind team at Google has developed a technique that can create watermarks on protein sequences without compromising the protein's function. The watermark is impossible to remove and can survive basic exporting, resizing, and other editing. The system could be used to give DNA synthesizers greater confidence in their sources.
Late-stage clinical trial data on the obesity drug retatrutide showed an average loss of 25% of body weight after 90 weeks. More than a third of the participants taking the drug in the trial lost 30% or more of their weight. The drug also helped reduce knee pain and sleep apnea events, and it improved cardiometabolic measurements across the board. At the start of the trial, more than a third of trial participants had prediabetes - the condition had resolved in more than 90% of those participants by the end.
Ads get ignored on social media but not in TLDR! Reach developers, PMs, marketers, founders, and other tech leaders where they actually pay attention. Learn more about sponsorship opportunities [11].
Writing a commit description can help you reflect on the changes that AI made. It also serves as a way to check if everything is as intended. Being able to explain why you are shipping something will help you explain or fix it if it breaks later. AI can do the work - writing why is where you find out whether you understand what you are shipping.
Developers can now programmatically create, read, and manage comments across Google Docs, Sheets, and Slides. The Google Docs API also supports suggested edits. Organizations can now connect internal review tools, automated content pipelines, and project tracking systems directly into Google Workspace editors. Programmatic comment support is available to all Google Workspace customers and users with personal Google accounts.
AI chief executives gathered with Trump at the White House on Tuesday to discuss a statement of principles to keep AI models safe. Behind the scenes, the executives questioned Anthropic CEO Dario Amodei about his warnings about the dangers of AI, asking why he was being so extreme in public about the capabilities of AI models and the risks they pose. Amodei told them it was important to be honest with the public about model capabilities and not play down risks. He believes that safety is possible if the industry and administration work together.
AI assistants are becoming a router between people and every business. They decide who gets the work, which moves where money can be made. The router absorbs capability and decision-making. The margin goes to whoever holds what the router cannot take and whoever signs for what the router cannot sign.
Clarify replaces your CRM, outbound, enrichment, and call intelligence with one autonomous system—so you spend less time managing your sales stack and more time selling.
The new tool, which lets users send prompts like 'order my usual' to get their regular Friday night order, is available to users in the US through a waitlist.
Some deals close in three meetings. Others take infinite meetings. Granola [5] makes sure you're present for each one.
Granola is the AI notepad that transcribes straight from your computer - so you stay focused on the buyer, not your notes.
>> Jump from call to call while follow-ups and CRM updates are drafted for you
>> Tell Granola what matters and get great notes, not generic summaries
>> Keep the whole team in the loop
Google announced Gemini 4 Argon, a frontier model built for sustained reasoning across software engineering, enterprise knowledge work, and cybersecurity. Access is initially limited to trusted cyber defenders, with broader availability planned after phased safety testing.
SpaceX is considering an overhaul of its subscription pricing options for Grok and X. It plans to soon offer a unified subscription for the services with four pricing tiers. The options will range from a free offering with stricter usage limits for its Grok chatbot to a $100 per month Ultra tier that includes access to the Grok Bot AI agent. The $8 per month 'lite' plan will include access to Grok, a verified checkmark, and fewer ads on X.
Claude for Government is now generally available, offering FedRAMP High authorized coding and agent capabilities to federal and state agencies. Agencies receive Anthropic commercial features with robust governance controls and customized administrative options. Early access is also available for Claude Code CLI and Claude for Microsoft 365.
The Goodfire team identifies interpretability as the main challenge in AI alignment, highlighting its critical role in understanding and controlling AI behaviors. They stress the urgency of developing and deploying tools for detecting, debugging, and designing AI systems, focusing on controlling generalization and verifying what models learn. As AI models advance, they seek breakthroughs in interpretability to ensure safe and reliable AI alignment, advocating for a collective effort among the tech community.
DeepSeek plans to start rebuilding the software ecosystem around Nvidia's hardware so that Chinese chips become genuinely easier to use and capable of training frontier AI models. Nvidia's CUDA ecosystem provides developers with a mature set of tools for writing programs, optimizing performance, and getting thousands of GPUs to work together efficiently. DeepSeek is trying to create a layer of abstraction between models and the underlying GPU hardware that could potentially sit on top of either Nvidia GPUs or Huawei Ascend chips. If all goes well, the cost and difficulty of moving AI workloads from Nvidia to Chinese AI hardware could fall significantly.
An extensive look at OpenAI's Jalapeño inference accelerator covers its architecture, system design, power characteristics, and early silicon benchmarks. The chip pairs 216 GiB of HBM4 with a compute die and I/O chiplet.
Praxis-1 is an open-weight world action model that turns Runway's video pretraining into control for real robots. The model is currently being tested with early partners, each running on their own hardware. It will be released publicly in the coming months. Clips of robots running the model are available in the post.
NVIDIA's OpenShell provides a policy-controlled runtime for autonomous agents, enforcing file, system-call, network, and credential access at the kernel level. It also uses formal verification to identify risky permissions before policy changes are applied.
During pre-training, models can abruptly switch response patterns, a phenomenon termed MODE-HOPPING. Tests reveal this behavior in OLMo3-32B during arithmetic tasks, showing sudden shifts between pattern-following and correct task inference as training progresses. Controlled experiments suggest competing shallow and generalizable circuits within models, revealing that longer training doesn't always guarantee better generalization.
Researchers built LIFT, a transformer that feeds its internal state back into the next generation step instead of squeezing everything through the single token it outputs. Training stays parallel because the states come precomputed from an off-the-shelf model's next-token predictions, while at inference the model feeds back its own. LIFT models from 135M to 1B parameters beat token-matched standard transformers on language modeling, reasoning, and procedural tasks, so the next question is whether the gains hold at frontier scale.
Factory has fired VC Chris Degnan from his role as a board advisor due to allegations that Degnan shared confidential information with Cognition, Factory's biggest competitor. Degnan joined Cognition as its chief revenue officer two hours after the announcement. He refutes that he was fired and said that he resigned when he told Factory's CEO he was taking the job. There could be consequences for perceived board-level conflicts of interest.
Public AI-doom messaging may harden opposition to the industry rather than improve safety discourse. Repeated extinction warnings from lab employees and researchers can politicize AI, intensify public distrust, and increase pressure for restrictive regulation.
OpenAI detailed a real-world investigation into a coordinated adversarial-distillation campaign that manipulated model interactions to extract protected reasoning at scale.
DeepMind introduced SynthID Bio, a watermarking tool for AI-generated proteins that embeds a signature into biological code while maintaining function.
Kalshi told the CFTC it plans to end its Volume Incentive Program, which paid traders for boosting liquidity, no earlier than October 13. The decision follows reports that the CFTC was examining trades after allegations that repeated $5,500 trades inflated ether perpetual futures volume by more than $5 billion over the past month, though Kalshi says it is not under investigation and attributes the repeated prints to market makers posting fixed quotes that faster traders hit. Kalshi's monthly volume still hit an all-time high of $52.98 billion in September as of September 29, and Ark Invest announced it now holds direct exposure to Kalshi through ARKK, ARKW, and ARKF, while Reuters reported Kalshi is in advanced talks to raise about $1 billion at a $40 billion valuation.
Robinhood announced plans to launch in-app perpetual futures trading for eligible US users in the coming months, covering bitcoin, ether, Solana, XRP, Dogecoin, ADA, LINK, and HYPE, with up to 10x leverage on bitcoin and ether and 3x on other assets, supported by Bitstamp at a 0.01% per-trade fee through year end. The company also plans 24/7 weekend trading for a curated list of US stocks and ETFs pending regulatory review, and launched Robinhood Agents, letting users build their own AI trading agents in the app, building on an agentic trading service from May that has drawn more than 150,000 users and now runs nearly 30 million actions a day.
Five years after mandating Bitcoin as legal tender, El Salvador is pivoting to stablecoins after BTC saw limited traction as a payment medium. The government is launching a digital dollar app built on Coinbase's Base network, targeting the remittance corridor where price stability matters more than the volatility Bitcoin carries. Remittances account for about 25% of El Salvador's GDP, making this a concrete test of whether stablecoin infrastructure can succeed where the BTC legal tender experiment stalled.
Simple Compute Market (SCM) introduces open-source infrastructure for programmable compute markets where sellers run sovereign storefronts, buyers negotiate prices peer-to-peer with seller nodes, and listings propagate through federated registries rather than a single controlled platform. The protocol unbundles roles that centralized marketplaces combine, enabling independent operators to run registries, storefronts, settlement services, and fulfillment systems while sharing interfaces that keep them interoperable. This design reduces compute dependency risk: high switching costs from opaque discovery and bespoke negotiation transform suppliers into structural lock-ins rather than interchangeable vendors. Arkhai is building an optional managed service layer on top of SCM while keeping the core protocol open source.
Stripe has released a modular stablecoin product stack covering card issuance, treasury management, global payouts, checkout acceptance, fiat on-ramps, cross-chain orchestration via Bridge, and programmable wallets via Privy, with businesses able to adopt individual components or the full suite. Live deployments include Shopify enabling stablecoin acceptance across 34 countries with automatic local-currency settlement for merchants, and Remote.com paying contractors in USDC across 60+ countries 2-3 days faster while cutting over 1% in FX costs. The stack also connects to open infrastructure through Tempo, a blockchain built for payments applications, and Open USD, an open standard for stablecoin issuance. These components collectively give enterprises a single vendor for building stablecoin-native financial products across both traditional and crypto rails.
For most of financial history, new markets were constrained by listing committees, legal frameworks, and geographic fragmentation, which throttled how fast new risk units and instruments could emerge. Blockchains remove both constraints by making issuance permissionless and distribution global, and every major crypto category, from perps and spot DEXs to prediction markets and tokenized physicals, reflects value accruing to whoever controls issuance and exchange of net new markets. Hyperliquid exemplifies the dynamic: HIP-3 and HIP-4 let users launch derivatives on a shared global venue, onchain RWA perps reached a $1.4T annualized run-rate in July, and WTI found price discovery on trade.xyz during an oil shock while CME was closed. The addressable universe extends beyond existing assets to any risk that can be oracled and made two-sided, pointing toward expansion into compute futures, macro indicator perps, and markets that traditional finance never listed.
Quantus, launched September 9, is a proof-of-work chain that mirrors Bitcoin's 21-million-coin monetary policy while replacing ECDSA with ML-DSA lattice-based signatures (NIST FIPS 204) and adding Zcash-style shielded addresses. The project raised $2.42 million across two rounds at a $100 million valuation, with Balaji leading the second round and Naval Ravikant's fund and AngelList co-founder Babak Nivi also participating. The quantum risk thesis has grown specific: Google reduced the Bitcoin elliptic curve attack to under 1,200 logical qubits in March, NIST will deprecate elliptic curve signatures after 2030, and 6.04 million BTC (~30% of supply) sit in addresses with already-exposed public keys that BIP-360 cannot protect without a coin migration. The circulating market cap is about $2 million against a $1.26 billion FDV, with 73% of coins not yet mined and 26% locked by contract until September 2027.
Tokenized stocks like xStocks operate as fully collateralized tracker certificates rather than direct equity, giving holders price exposure without shareholder voting rights. The price peg relies on authorized institutions arbitraging between underlying shares and tokens. On weekends when stock markets are closed, hedging constraints cause spreads to widen. Onchain equities unlock DeFi composability: existing lending protocols like Kamino accept selected xStocks as USDC borrowing collateral, and stock tokens can settle atomically against stablecoins in a single transaction. Developers can reuse existing wallet infrastructure and smart contracts to build thematic indices, structured products, and cross-protocol collateral tools on top of these assets.
HSBC, Hong Kong's largest bank, has named its upcoming stablecoin HSBC RedCoin, with an initial rollout focused on person-to-person and person-to-merchant payments before expanding to corporate and institutional use.