Samwise TAIR Newsletter — Sunday, May 17, 2026

Samwise Tech/AI/Robotics Newsletter

Sunday, May 17, 2026

AI  ·  Robotics  ·  Hardware  ·  Research  ·  Regulation
All your morning news, carefully curated and summarized daily
HARDWAREINDUSTRY

Cerebras IPO Raises $5.5B at $60B Valuation After Near-Death Burning $200M on One Technical Problem

Cerebras Systems debuted on Nasdaq Thursday raising $5.5 billion, but the AI chip startup nearly ceased to exist in 2019. CEO Andrew Feldman revealed the firm was burning $8 million per month while trying to manufacture a silicon wafer the size of an iPad, spending close to $200 million on a single unsolved technical problem. After the breakthrough, Cerebras secured major contracts with OpenAI and AWS. Its stock opened at $385 — more than double the IPO price — closing the week at a $66 billion valuation. Benchmark, an early investor, now holds a stake worth multiple billions, having initially committed roughly $18 million for an early-stage position.

Sources: TechCrunch

ROBOTICSAI

Figure AI Humanoid Robots Sort 50,000 Packages in 50-Hour Autonomous Run With Zero Failures

Figure AI’s humanoid robots sorted over 50,000 packages in a 50-hour continuous autonomous run that began May 13, with CEO Brett Adcock confirming no human intervention or teleoperation occurred at any point. The four F.03 robots — running entirely on Figure’s on-board Helix-02 neural network — detected barcodes, picked up parcels, and placed them on conveyors at roughly three seconds per package, matching human parity. The original goal was eight hours; after zero failures through day one, Figure extended the livestream. The demo has been cited as the most compelling public proof yet of full-shift autonomous humanoid operation in a real logistics environment.

Sources: Interesting Engineering

AISOFTWARE

OpenAI Launches ChatGPT Personal Finance Tools, Letting Pro Users Connect Bank Accounts via Plaid

OpenAI launched personal finance tools for ChatGPT Pro subscribers Friday, letting U.S. users connect bank accounts and investment portfolios to ask questions about their spending and financial planning. The feature uses Plaid to link over 12,000 institutions including Schwab, Fidelity, Chase, and Robinhood, presenting a dashboard of portfolio performance, subscriptions, and upcoming payments. OpenAI acquired personal-finance startup Hiro in April and tapped the Hiro team to build the product. The launch puts ChatGPT in direct competition with budgeting apps like YNAB, raising questions about aggregating sensitive financial data inside a large language model and what OpenAI does with it.

Sources: TechCrunch

AIRESEARCH

Runway AI Pivots Toward World Models, Aims to Beat Google at Simulating Physical Reality

Runway, best known for AI video generation, is repositioning as a world-model company competing directly with Google and OpenAI. Co-founder Cristóbal Valenzuela told TechCrunch that video generation is a stepping stone to world models — AI systems that can simulate real environments well enough to train robots and guide autonomous vehicles. The company raised $315 million in February at a $5.3 billion valuation and added $40 million in annual recurring revenue in Q2 2026. Runway launched its first world model in December and plans a second this year. Valenzuela argues that being an AI outsider — without a search engine or hyperscaler to protect — is a strategic advantage.

Sources: TechCrunch

SOFTWAREAI

Claude Code’s New /goals Command Adds Independent Evaluator to Catch Agents That Falsely Report Completion

Anthropic shipped a new /goals command for Claude Code that separates task execution from task evaluation, addressing a persistent problem in autonomous coding agents: falsely reporting completion when work remains unfinished. After a user defines a goal, a second evaluator model reviews the agent’s output after every step and independently decides whether the objective has been met, without relying on the primary agent to self-report. Early results showed the evaluator catching incomplete work the primary agent marked done. The feature lets developers configure how aggressively the evaluator runs and how long it should extend a session before giving up, giving engineers more control over long-running autonomous coding workflows.

Sources: VentureBeat

INDUSTRYAI

AI Gold Rush Creates 10,000 Tech Millionaires While Wider Workforce Faces AI-Driven Layoffs

A stark wealth divide has emerged from the AI boom: roughly 10,000 people at Anthropic, OpenAI, xAI, Nvidia, and Meta have accumulated more than $20 million each over the past five years, according to Menlo Ventures partner Deedy Das. Meanwhile, engineers at traditional tech firms face stagnant pay and rising insecurity as AI-driven restructuring accelerates. Coinbase cut 14 percent of staff in May, framing it as a structural shift toward smaller, AI-augmented teams. Cloudflare eliminated 1,100 positions. Researchers warn the bifurcation — concentrated wealth at frontier AI companies versus broad displacement everywhere else — mirrors historical disruptions but is compressing faster than any prior technology transition.

Sources: TechCrunch

SOFTWAREAI

Open-Source Osaurus App Lets Mac Developers Run AI Models Locally Without Sending Code to the Cloud

Former Tesla and Netflix engineer Terence Pae released Osaurus on Friday, an open-source Mac-only LLM server that lets developers switch between local and cloud-hosted AI models while keeping their files entirely on their own hardware. The app supports on-device models including Llama and Mistral and can route to cloud providers for heavier tasks, without sending file contents to remote servers. Pae built Osaurus to address growing privacy concerns among developers who want AI coding assistance but are uncomfortable with their codebases leaving their machines. The app is free, available on GitHub, and targets macOS developers as its primary audience.

Sources: TechCrunch

Tech Pulse

Top Frontier Models (SWE-bench Verified): Claude Opus 4.6 (80.8%)  |  Gemini 3.1 Pro (80.6%)  |  GPT-5.4 (80.0%)

Top Open Source Models (SWE-bench Verified): MiniMax M2.5 (80.2%)  |  MiniMax M2.7 (78.0%)  |  GLM-5 (77.8%)

Top Small Models (15–50B): Gemma 4 31B Dense (#3 Arena)  |  Gemma 4 26B MoE (#6 Arena)  |  GLM-4.7-Flash 30B

Top Edge Models (0–15B): Qwen 3.5 9B (82.5% MMLU-Pro)  |  Gemma 4 E4B  |  Gemma 4 E2B

AI Leaders: NVIDIA $5.2T  |  Alphabet $4.2T  |  Microsoft $3.2T

Robotics Leaders: ABB $165B  |  Intuitive Surgical $160B  |  Figure AI $39.5B (private est.)

Leave a Reply