It’s Friday, and you know what that means—a fresh edition of Insane AI is here. Ready to dive in?
1: OpenAI debuts GPT-4.5, its biggest model yet
OpenAI has launched GPT-4.5—a new, larger language model that it says is better at recognizing patterns, avoiding “hallucinations,” and delivering more natural conversations. The upgrade is notable because it relies on massive “pre-training,” at a time when AI companies are shifting their focus toward “reasoning” (letting models take more time to process queries).
What’s new
GPT-4.5 is initially available to ChatGPT Pro users paying $200 per month, as well as API customers. OpenAI plans to roll it out soon to ChatGPT Plus subscribers ($20/month) and enterprise customers.
It should handle real-time search and file uploads, though voice mode, video, and screen sharing won’t arrive until later.
Compared with GPT-4o (OpenAI’s more budget-friendly model), GPT-4.5 is pricier—30 times as much for developers—and more resource-intensive to train and run.
Why it matters: CEO Sam Altman calls GPT-4.5 the first model that “feels like talking to a thoughtful person,” hinting it’s a leap rather than a small step. But the bigger-is-better approach has become hugely expensive, with diminishing returns in pure pre-training. Future models, including GPT-5, will blend both large data training and deeper “chain-of-thought” reasoning to deliver more precise answers.
What’s next: OpenAI says GPT-4.5 is its final model that doesn’t heavily emphasize reasoning. The company aims to merge these larger, more knowledgeable models with “thinking” capabilities—making future versions even more advanced at logic, math, and big data tasks. For now, GPT-4.5 addresses a growing demand for high-stakes enterprise use cases, where accuracy matters more than cost.
Learn more about GPT-4.5 here.
The HubSpot Developer Platform is a trusted way for thousands of organizations to connect new and existing software. With flexible UI and extensibility tools, you can build apps that help teams unify their tech stack across our platforms. From listing your app in the HubSpot Marketplace to building custom solutions, there’s something for everyone to grow better.
2: CFOs crash the AI buying party: Focusing on ROI over hype
Chief financial officers are taking center stage in AI purchasing decisions. According to Writer co-founder Waseem Alshikh, CFOs now weigh in on more than 70% of new AI deals, up from less than 5% a year ago. That shift is pushing vendors to prove a clear return on investment rather than simply tout the latest model architectures.
Rather than chasing splashy “Chief AI Officer” hires or fiddling with technical specs, companies want AI tools that cut costs and improve worker productivity. In many cases, CFOs are redirecting funds from hiring toward AI systems they hope will do more with fewer people. It’s a far cry from the bolder visions of AI-fueled revenue growth that dominated headlines two years ago—instead, it’s all about the bottom line.
3: Upfront Summit’s wake-up call: AI’s bubble may burst
This year’s Upfront Summit in Los Angeles brought together VCs, limited partners, and a wave of emerging fund managers. While everyone was talking about raising new funds—“always be raising”—investors also sounded a note of caution about artificial intelligence. Here are the takeaways:
Everyone’s fundraising. From established firms to newcomers, it felt like nearly every attendee was hunting for fresh capital. Chelsea Clinton, for instance, shared plans to raise a third fund for Metrodora Ventures next year, aiming for $50 million. There also seemed to be more first-time or “solo” fund managers than ever before. Some insiders speculated that starting a fund is the new “soft landing” for investors leaving larger firms.
AI “greed” and the risk of down rounds. Prominent voices like Vinod Khosla (an early OpenAI backer) described a “great greed phase” in AI investing, predicting that 80% of current AI startups will fail to generate returns within five years. Sarah Guo of Conviction said there won’t be room for “infinite” $100B AI companies. While some ventures will win big, plenty of investors will lose money when the hype subsides.
Sky-high valuations, big future questions. Anthropic recently closed funding at a $60B+ valuation, and OpenAI is rumored to be raising at a jaw-dropping $300B. Even brand-new AI startups are commanding massive valuations—Mira Murati’s Thinking Machines Lab is reportedly aiming for $9B, before generating any revenue. Right now, there’s little fear among big investors, but as the AI frenzy continues, many expect a “correction” to rein in overblown valuations.
At Upfront Summit, it was clear that fundraising—whether for a new VC fund or an AI startup—is still going strong. However, seasoned investors warn that AI is headed for a reality check, with more down rounds likely on the horizon. In other words, the money may be flowing freely now, but not everyone will come out a winner in the end.
4: Amazon cracks down on AI cheating in job interviews
Amazon is barring job applicants from using AI tools like ChatGPT during interviews, new internal guidelines show. If recruiters suspect a candidate is relying on AI “teleprompter” apps or coding assistants, that applicant can be disqualified. Indicators might include a candidate typing mid-response, reading scripted lines out loud, or seeming confused by AI-generated mistakes.
Some Amazon employees see these tools as dishonest, while others argue they can raise quality, especially for technical roles. A recent viral video of someone claiming they got an Amazon offer by secretly using AI sparked the crackdown. Though Amazon is testing ways to let workers use generative AI internally, the company insists job seekers should demonstrate genuine skills in their interviews. As AI becomes more common, big employers face a growing dilemma: how to spot—and handle—“augmented” candidates.
Editors picks ☕️
Reuters: AI robots may hold key to nursing Japan's ageing population.
Amazon: Introducing Alexa+, the next generation of Alexa.
The Verge: Meta’s AI chatbot will soon have a standalone app.
