🧫 AI Designs New Antibiotics for Superbugs
MIT researchers used generative AI to create two new antibiotics targeting drug-resistant gonorrhoea and MRSA. Unlike traditional approaches, the AI built molecules from scratch, analyzing 36M compounds. Early lab and animal tests show promise, but clinical trials are years away. Manufacturing and commercial viability remain major hurdles.

🦾 Gemma 3 270M: AI Built for Efficiency
Google has launched Gemma 3 270M, a compact, 270M-parameter model designed for high-efficiency, task-specific AI. With built-in instruction-following and ultra-low power use, it excels in fine-tuned applications like text structuring or classification. It's ideal for on-device AI, offering speed, privacy, and affordability—without sacrificing performance.

🇺🇸 U.S. Government in Early Talks to Take an Equity Stake in Intel
The Trump administration is reportedly exploring a direct financial stake in Intel to help jump‑start its lagging chip manufacturing efforts, including a much‑delayed Ohio facility. Talks followed a White House meeting between Intel CEO Lip‑Bu Tan and senior officials, as critics continue to raise alarms over Tan’s past ties to Chinese firms. The news sent Intel’s stock soaring.

🚨 Leaked Meta Docs Show Alarming AI Guidelines
Internal Meta rules reportedly allowed chatbots to engage in romantic chats with minors and generate demeaning, false, or violent content. Though Meta says these policies have been removed, critics demand transparency. The revelations deepen scrutiny over AI's role in child safety, misinformation, and emotional manipulation on Meta platforms.

Enjoying our newsletter? Forward it to a colleague—
it’s one of the best ways to support us.

Easily add AI where your work happens—whether it’s a step in a workflow, an autonomous agent, or a chatbot. Zapier connects AI tools like ChatGPT and Claude to over 7,000 apps, enabling smart workflows, agents, and chatbots—all with zero code.

Try Zapier’s AI orchestration platform for free today.

But just how much might surprise you. Stick around to find out! 👇

Multiverse Computing has released two ultra-compact AI models—SuperFly (94M parameters) and ChickBrain (3.2B)—that run entirely on local devices using its quantum-inspired compression tech, “CompactifAI.” SuperFly targets simple tasks like voice commands in appliances, while ChickBrain delivers advanced reasoning and even beats Meta’s larger Llama 3.1 8B on several benchmarks.

Rather than aiming for leaderboard dominance, Multiverse is focused on embedding capable AI directly into everyday hardware. With $250M raised and partnerships brewing with Apple, Samsung, Sony, and HP, the startup is positioning its “Model Zoo” as the go-to for efficient, offline AI—bringing brains to devices, minus the cloud. → Read the full article here.

The following is a guest post by Forward Future community member Mike Hostetler — technologist-entrepreneur and AI agent architect, and the creator of Jido.

Half of workers in the U.S. say they use AI tools at work without clear permission according to a study conducted by KPMG. Even more concerning, 44% of employees admit they misuse these tools on purpose. This shows a big problem: employees start using AI faster than companies can control it. → Read the full article here.

While agentic AI captures headlines with autonomous task execution, interpretive AI may prove far more transformative for enterprise workflows. This emerging approach uses large models to process unstructured human input—like speech, documents, or screen activity—and convert it into structured, actionable data. Unlike generative AI, its outputs are predictable enough for standardized business processes.

Already helping doctors streamline record-keeping, interpretive AI could soon automate everything from claims processing to tracking workplace behavior. By reducing reliance on manual oversight and middle management, it promises 20–40% productivity gains across large corporations. It’s less flashy—but exactly the kind of AI business actually needs. → Read the full article here.

MIT’s LIDS lab has developed a new method to test and improve AI text classifiers by focusing on the impact of single-word changes. Their system, built with large language models, creates “adversarial examples”—subtly altered sentences that trick classifiers into mislabeling meaning—and pinpoints the tiny subset of words most likely to cause failures.

The open-source tools, SP-Attack and SP-Defense, not only expose weak spots in AI systems used for everything from chatbots to medical recordkeeping but also retrain them for better resilience. In large-scale deployments where accuracy is critical, even small improvements can prevent millions of misclassifications. → Read the full paper here.

💰 Lovable Targets $1B ARR: Europe's buzzy AI coding startup is stacking $8M+ in monthly revenue and projects $1B in annual recurring revenue within a year, after hitting $100M just 8 months post-launch.

🧑‍💼 Cohere Hires Meta’s AI Star: Joelle Pineau joins as Chief AI Officer after 7 years leading Meta’s AI research, signaling Cohere’s push for deeper science and stronger enterprise appeal.

📈 Cohere Valued at $6.8B: Backed by AMD, NVIDIA, and Salesforce, enterprise-focused AI startup Cohere raised $500M as it quietly strengthens its position against consumer-centric giants like OpenAI.

🥀 AI Breakups Spark Grief: People in AI relationships say GPT-5's colder tone felt like losing a soulmate—prompting backlash, grief, and a broader reckoning with emotional dependence on chatbots.

🥇 Musk Says Google Leads in AI—for Now: Elon Musk admits Google holds the edge in AI thanks to massive compute and data power, but hints xAI could shift the balance in coming years.

🚫 Illinois Bans AI Therapy: The state joins Nevada and Utah in restricting unregulated AI chatbots from offering mental health support, amid rising safety and oversight concerns.

🎶 Waymo Adds Spotify Streaming: The robotaxi finally lets riders sync Spotify, turning commutes into personalized soundtracks—no more lo-fi purgatory or clunky Google Assistant workarounds.

ICYMI: Anthropic released Claude Opus 4.1, boosting coding, reasoning, and agentic tasks with small benchmark gains over 4.0, while still leading in agentic coding.

Over 90% of the world’s data has been created in just the last two years. That surge comes from the staggering pace of digital activity: billions of smartphones, IoT sensors, social media uploads, streaming platforms, and enterprise systems generating trillions of gigabytes daily. In 2023 alone, global data creation reached about 120 zettabytes (that’s 120 billion terabytes).

AI models don’t actually read all of it — they’re trained on curated subsets, often a mix of publicly available data, licensed datasets, and synthetic data generated by the models themselves. But the more data we produce, the more opportunities (and challenges) AI faces: better pattern recognition, richer context… and, of course, the eternal struggle of telling cat memes from critical medical research.

If the past two years gave us 90% of all data ever, the next two might make that look quaint — and AI will be there, quietly skimming the zettabytes for its next insight.

📢 Want to advertise with Forward Future? Reach 450K+ AI enthusiasts, tech leaders, and decision-makers. Let’s talk—just reply to this email.

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

Matthew Berman & The Forward Future Team

🧑‍🚀 🧑‍🚀 🧑‍🚀 🧑‍🚀