Presented by
Good morning, AI enthusiasts. OpenAI's GPT-5.5 just ran 150+ hours of protein folding research with no one watching, no prompts in between, and results to show for it.
Which raises a question: if AI can now run extended scientific experiments without supervision, what does a research lab's daily schedule actually look like?
In today's recap:
GPT-5.5 runs 150+ hours of solo science
OpenAI connects your bank to ChatGPT
Map a competitor's full traffic with Manus
Vatican launches its first AI ethics group
3 new AI tools, prompts, and more
OPENAI
GPT-5.5 runs 150+ hours of science alone
Recaply: OpenAI's GPT-5.5 just demonstrated autonomous scientific research capability, with community users reporting the model running 150+ hours improving protein folding models without human supervision and OpenAI's own benchmarks showing a 32% jump in scientific research performance.
Key details:
GPT-5.5 operates as a fully agentic system, planning multi-step research tasks, checking its own work, and iterating over long timeframes with no human intervention between steps.
The model scores 25.0% on GeneBench for scientific tasks, up from GPT-5.4's 19.0%, while hitting 82.7% on Terminal-Bench 2.0, the highest score among frontier models tested.
According to OpenAI, GPT-5.5 "uses significantly fewer tokens to complete the same tasks," meaning long autonomous runs cost less compute than earlier models of comparable scope.
GPT-5.5 launched April 23, 2026 for Plus, Pro, Business, and Enterprise users in ChatGPT and Codex, with API access added the following day for partners and customers.
Why it matters: There's been plenty of talk about AI in science, but most of it meant AI assisting with analysis, not running experiments on its own. GPT-5.5 spending 150+ hours on protein folding without human supervision changes that. It doesn't just mean faster research. It means work running overnight, on weekends, with no one checking in. Labs that set up these workflows won't just publish more papers, they'll operate at a pace most research teams can't match.
PRESENTED BY HIGHTOUCH
Scale winning creative without sacrificing quality
Ad algorithms reward volume, but most teams can’t keep up without sacrificing either quality or brand. Hightouch changes that.
With Ad Studio, performance marketers can turn their data into hundreds of high-quality ads in minutes. Every asset is grounded in real insights, built for easy editing, and ready to launch — across any platform.
React to performance in real time. Jump on cultural trends while they’re still relevant. Respond to competitor campaigns before they gain traction.
More output, better inputs, no trade-offs required.
OPENAI
OpenAI brings personal finance into ChatGPT
OAI
Recaply: OpenAI just rolled out a personal finance preview in ChatGPT for Pro users in the U.S., letting users connect financial accounts, view spending dashboards, and ask questions grounded in their real financial data.
Key details:
Users connect financial accounts through Plaid, with support for more than 12,000 institutions, allowing ChatGPT to sync and categorize spending data alongside context like savings goals or upcoming purchases shared in conversation.
More than 200 million people already use ChatGPT monthly for finance questions, with the new feature drawing on GPT-5.5's improved reasoning for the complex, context-dependent decisions personal finance often requires.
Beyond connected accounts, ChatGPT saves user details to "Financial memories," meaning mortgage balances, savings goals, and loan history persist across future conversations, according to OpenAI.
Available now as a preview to Pro users in the U.S. on web and iOS, with Plus expansion next and the goal of making it available to all users eventually.
Why it matters: Personal finance apps have always put the work on users. Categorize transactions, set budgets, check in every week. ChatGPT connected to your actual bank can answer "can I afford to buy a car in six months?" with your real numbers. The Financial memories feature is the sleeper here. A ChatGPT that remembers your mortgage, savings goals, and loan history isn't just a calculator, it starts to look like a standing financial advisor that doesn't charge by the hour.
GUIDES
Reverse-engineer a competitor's traffic with Manus and Similarweb
Recaply: In this tutorial, you will learn how to use Manus and its Similarweb integration to map a competitor's full digital growth strategy, including top keywords, referral sources, and high-traffic pages, in under 10 minutes.
Step-by-step:
Open Manus and start a task with your competitor's domain. Ask: "Use Similarweb data to analyze [competitor]. Show traffic trends for the last 6 months, top channels, and how organic vs. paid breaks down."
Drill into keywords by asking: "Which search terms drive the most traffic to [competitor]? Break down branded vs. unbranded demand and flag the top 10 by difficulty."
Map their referral sources: "Show the top sites sending traffic to [competitor] and where visitors go after they leave." This uncovers affiliates, media coverage, and community channels your competitor relies on.
Ask for a multi-competitor brief: "Compare [competitor] with its top 3 rivals using Similarweb data. Include traffic trends, keywords, incoming referrals, and most-visited pages. Format as an executive brief."
Export the output in your preferred format by asking Manus to structure findings as a slide deck, spreadsheet, or shareable PDF, with charts for traffic trends and channel breakdowns.
Pro tip: For the richest single output, use this prompt from Manus's own docs: "Use Similarweb data to build a competitive brief on [competitor]. Cover the 6-month traffic trend, top growth channels, branded vs. unbranded SEO mix, referral sources, and most-visited pages. Package as a slide deck."
TOGETHER WITH WISPR FLOW
Stop re-prompting. Say it right the first time.
Voice-first prompts preserve the nuance you cut when typing. Speak once, paste into any AI tool, get results that don't need a follow-up. 89% of messages sent with zero edits.
TOOLS
Trending AI Tools
🎨 Ardot - Tencent's AI-native design agent platform covering the full UI/UX workflow, from prompt-to-design to one-click design-to-code with Figma import and real-time collaboration
💻 ScreenCharm - macOS screen recorder with built-in smart zoom, motion blur, and MacBook-style frames, $79 lifetime license
🤖 Gemini 3.2 Flash - Google's next-gen Flash model spotted in Antigravity, with early testers reporting it outperforms Gemini 3.1 Pro on coding and animation tasks
NEWS
What Matters in AI Right Now?
Former Google CEO Eric Schmidt drew sustained boos from University of Arizona graduates during his commencement address when he praised AI, with students interrupting as he acknowledged their "rational" fear that "the machines are coming, that the jobs are evaporating." Multiple speakers at separate schools were booed for similar AI comments the same week.
Manus deepened its Similarweb integration for Pro users, adding keyword demand, referral sources, landing page performance, and competitor organic traffic data to its conversational AI workflows, with usage billed at 17 credits per keyword query.
Perplexity launched a Snowflake connector for Computer, letting enterprise teams query live warehouse data in plain language and receive SQL-backed answers on product usage, customer segments, and campaigns, with admins retaining role-based access controls.
Pope Leo XIV created a Vatican study group on artificial intelligence ahead of his first encyclical, expected to call for an ethics-based approach prioritizing human dignity, with the announcement coming 135 years to the day after Pope Leo XIII dated "Rerum Novarum."
Vercel Labs released Zero, an experimental systems programming language built for AI agents, with the compiler emitting structured JSON diagnostics that agents can parse and repair directly rather than interpreting human-readable error messages, earning 1.9k GitHub stars since May 15.
More than 60 MAGA allies including Steve Bannon sent a letter to President Trump urging mandatory government testing of powerful AI models before release, comparing frontier AI to nuclear systems and aviation safety, putting them at odds with the White House's current hands-off approach.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
