📈 OpenAI Eyes $1T IPO by 2027
OpenAI is laying early plans for a public listing that could value it at up to $1 trillion, sources told Reuters. A filing could come as soon as late 2026, driven by surging revenue and reduced reliance on Microsoft. The IPO would fund massive AI infrastructure investments as demand grows.

🤔 Anthropic Tests Early Signs of Model Introspection
Anthropic reported that its Claude models show limited “introspection,” occasionally detecting and describing their own internal activations using a “concept injection” method. In tests, Opus 4 and 4.1 recognized injected concepts about 20% of the time, an unreliable but intriguing step toward AI transparency and model self-awareness.

🪢 Figma Launches AI Workflow Tool 'Weave'
Figma has acquired Weavy and rebranded it as Figma Weave, a new tool that lets users visually connect multiple AI models and editing tools on a single, node-based canvas. Weave allows creators to chain tasks across models or compare outputs simultaneously—without switching apps. It's aimed at designers who want more control over generative workflows.

🚀 Canva Debuts Design AI Model, Expands Creative Tools
Canva has launched its first proprietary design model, trained on its own elements to generate editable, layered visuals across formats like presentations, whiteboards, and social posts. New features include smarter AI assistants, 3D object generation, and a spreadsheet-linked widget builder. Canva also unveiled “Canva Grow,” a marketing platform powered by its creation tools.

Multimodal AI can generate photorealistic images, write poetry, and solve complex math problems, but can it read an analog clock? Or tell you what day July 4, 1776, fell on?

Answer below! 👇

Lindy turns an expensive marketing agency into a powerful set of AI agents any marketing team can use to bring high-performing campaigns to market faster. Study the competition. Spot opportunities. Write messaging. Create campaigns. Then do it on repeat, faster than any human could.

Start building your AI workforce at lindy.ai, the easiest way to build AI agents.

Humanoid robots hit the market, new thermodynamic chips promise 10,000× efficiency, open-source AIs rival GPT-5, and Tesla plans a distributed supercomputer.

Guest lineup:

Join us today at 10 AM PT. Streaming live on YouTube or X!

OpenAI has launched Aardvark, an autonomous AI agent designed to identify and patch security vulnerabilities in software codebases. Now in private beta and powered by GPT-5, Aardvark scans repositories, models threats, validates potential exploits in a sandbox, and suggests Codex-generated patches.

In internal and partner testing, it detected 92% of known and synthetic bugs, surfacing vulnerabilities missed by conventional tools. Why it matters: As software security threats rise—over 40,000 CVEs were logged in 2024—Aardvark aims to scale security expertise and speed up response without disrupting development. → Read the full article here.

Amazon, Salesforce, Meta, and other major firms have announced over 50,000 job cuts in recent months, often citing artificial intelligence as the driver. Amazon alone is eliminating 14,000 roles, calling AI a “transformative” force—though later downplayed its role in the cuts.

Despite rising AI investments, recent studies show mixed returns: 60% of firms reported minimal financial gains, and only 10% saw meaningful ROI from agentic AI. Why it matters: Experts warn AI may be serving as a convenient cover for old-fashioned cost-cutting amid economic headwinds and investor pressure. → Read the full article here.

Lindy turns an expensive marketing agency into a powerful set of AI agents any marketing team can use to bring high-performing campaigns to market faster. Study the competition. Spot opportunities. Write messaging. Create campaigns. Then do it on repeat, faster than any human could.

Start building your AI workforce at lindy.ai, the easiest way to build AI agents.

💸 Microsoft Books $3.1B OpenAI Hit: A $3.1 billion drop in Q1 net income reflects Microsoft’s $13B OpenAI stake, still, Azure revenue jumped 40% as the cloud-AI duo evolves into rivals.

▶️ YouTube Offers Buyouts in AI Shift: CEO Neal Mohan announced voluntary exits as YouTube restructures into three AI-driven units—viewer, creator, and subscription—to boost speed and innovation.

📉 Meta Shares Fall on AI Costs, Tax Hit: A $15.9B tax charge and soaring AI spend dragged Meta’s Q3 EPS to $1.05, far below forecasts, fueling investor fears of an overheating AI market.

🙅 Trump Skips Blackwell Chip in Xi Talks: Despite hinting otherwise, Trump didn’t raise NVIDIA’s AI chip with Xi Jinping—sidestepping a flashpoint as U.S. export curbs strain tech ties with China.

🤥 Illinois Students Admit to AI Cheating (Paywall): Dozens at the University of Illinois confessed to using ChatGPT for assignments, exposing gaps in campus policies as schools race to adapt to generative AI.

Salesforce: Agentforce/AI Deployment Strategist

Apple: Sr. Manager, Internal Tools & Operations

Cisco: Senior Manager, Machine Learning Engineer

Despite all the hype around advanced multimodal AI models, it turns out they're still pretty bad at everyday tasks you probably mastered in elementary school. Recent research shows these systems correctly interpret analog clocks only 38.7% of the time. Want them to tell you what day of the week a date falls on? Good luck—accuracy drops to just 26.3%.

Why the struggle? These tasks require nuanced reasoning across visual and temporal domains—something current models aren’t quite wired for yet. So next time someone tells you AI is ready to run the world, maybe ask it to read a clock first.

📥 Got a hot tip or burning question? Drop us a note! The best reader insights, questions, and scoops may be featured in future editions. Submit here.

Thanks for reading today’s newsletter—see you next time!

Matthew Berman & The Forward Future Team

🧑‍🚀 🧑‍🚀 🧑‍🚀 🧑‍🚀