Good morning. It's Friday, April 24, and we're covering Musk’s high-stakes OpenAI trial, Google’s breakthrough in distributed AI training, GPT-5.5’s push toward more autonomous models, and more.

We're hiring at Forward Future. If you live and breathe AI and want to help shape how millions of people understand it, we want to talk. More details here.

OpenAI GPT-5.5 Pushes Autonomy While Holding Latency Steady
Rather than slowing down with added capability, GPT‑5.5 matches prior latency while increasing autonomy and efficiency. OpenAI introduces the model as a step toward agentic AI that can plan, use tools, and complete multi-step tasks. Benchmarks show gains like 82.7% on Terminal-Bench and higher scores across coding and math evals, often with fewer tokens. The release rolls out to ChatGPT tiers and Codex, with API access pending.

Pentagon’s $54 Billion Surge Toward Autonomous Warfare
Rather than incremental defense upgrades, the Pentagon proposes a 24,000% funding jump for autonomous warfare systems. The $54 billion 2027 budget request centers on a new Defense Autonomous Warfare Group to expand drone and AI-powered military capabilities. Funds target autonomous systems across air, land, sea, and subsea operations, with programs like “Drone Dominance” and partnerships with frontier AI labs. Experts warn the U.S. lacks doctrine and safeguards for deploying such technologies at scale.

Tesla Boosts Spending As AI Pivot Pressures Cash Flow
Rather than tightening costs, Tesla triples capital spending to $25 billion as it shifts toward AI and robotics. The 2026 plan expands investments in compute infrastructure, chip design, manufacturing, and robotaxi and Optimus programs under Elon Musk. The figure jumps from $8.5 billion in 2025, with added funding for data centers, a semiconductor fab in Austin, and new robot production lines. CFO Vaibhav Taneja says the push will drive negative free cash flow later in 2026 despite $44.7 billion in reserves.

Mozilla Uses AI Model To Surface Hundreds Of Hidden Bugs
Rather than relying on human researchers, Mozilla finds 271 Firefox vulnerabilities using an AI model scanning unreleased code. The issues were identified by Anthropic’s Mythos Preview and fixed in Firefox 150, marking a shift toward automated security analysis. The model far outperformed earlier systems, which flagged just 22 bugs, and can sweep large codebases faster and cheaper than manual audits. Executives warn the approach could widen gaps between well-funded firms and open-source projects.

FRIDAY FACTS

In 2017, 8 Google researchers published the paper that made every modern LLM possible — GPT-5.5, Claude, Gemini, all of it. Guess how many still work at Google.

Accio Work is an AI-powered workspace for business owners, solopreneurs, founders, operators, and modern businesses looking to move faster. It manages sourcing, supplier negotiation, store operations, and marketing end to end.

Powered by Alibaba.com’s 1B+ products and global supplier data, plus built-in tools, it executes tasks while you stay in control. From idea to execution, everything runs in one seamless workflow with no setup or complexity, just faster, smarter, scalable results.

Try Accio Work now — 7 Days Free!

OpenAI isn’t just going for raw intelligence. They’ve improved the personality of the model. This is almost certainly to capture more of the personal agent (OpenClaw) market. Its responses are shorter, more human-like, and less formal. It actually has a personality. → Read the full article here.

A jury trial beginning Monday in Oakland, California will determine whether Elon Musk’s 2024 lawsuit against OpenAI and CEO Sam Altman—seeking more than $150 billion in damages—has merit. Musk alleges OpenAI abandoned its founding nonprofit mission and misused his early financial backing, while OpenAI counters that Musk himself pushed for a for-profit structure before leaving in 2018.

The case will feature testimony from key figures including Satya Nadella and Mira Murati and could last several weeks. Its outcome could reshape the AI industry, potentially disrupting OpenAI’s growth, partnerships with Microsoft, and a possible future IPO, while influencing competitors such as Google and Anthropic. → Read the full article here. (Paywall)

Google researchers unveiled Decoupled DiLoCo, a new distributed training architecture designed to train large AI models across geographically dispersed data centers with lower bandwidth and higher fault tolerance. The system splits training into asynchronous “islands” of compute, allowing failures in one region to avoid disrupting others while maintaining performance.

In tests, the method achieved up to a 16-fold reduction in communication bandwidth compared to baseline distributed training methods. The approach also enabled training a 12-billion-parameter model across four U.S. regions using standard network infrastructure, achieving comparable accuracy while running over 20 times faster than conventional techniques. → Read the full article here.

Matt critiques Anthropic’s pricing, compute limits, and poor communication—arguing missteps are pushing users to OpenAI amid surging AI demand.

Era Raises $11M for AI Device Platform: Startup builds software layer enabling AI-powered gadgets across 130+ models and 14 providers, aiming to replace apps and support scalable, customizable hardware ecosystems.

Anthropic Targets Europe Data Centers: AI firm hires six-figure dealmaker to secure compute capacity across key hubs, expanding infrastructure amid $600 billion hyperscaler spending surge and intensifying global AI buildout.

Tencent Open-Sources Hy3 AI Model: New 295B-parameter Mixture-of-Experts system with 21B active parameters and 256K context boosts reasoning, coding, and agent tasks while cutting latency and inference costs.

Kling AI Launches 4K Video Mode: Video 3.0 update enables one-click 4K generation with improved visual consistency, detail, and lighting, offering up to 20% discounts for subscribers through May 7.

FRIDAY FACTS

They went on to found or co-found Character.AI, Cohere, Adept, and Essential AI, companies now worth a combined $25B+, all competing against Google. The paper was "Attention Is All You Need," and it introduced the transformer.

Thanks for reading. See you next time!

— Matthew Berman, Nick Wentz & the Forward Future Team