⏳ The AI Slowdown

The Street

Is AI Slowing Down? Meta’s Delay Says Yes

Meta’s new AI model, Behemoth, was supposed to launch in April—now it’s delayed until fall or later.

Across the board, major labs are slowing down: OpenAI’s GPT-5 has almost reached a year in delays, and Anthropic’s Claude 3.5 Opus is still waiting in the wings. Even Google’s Gemini 2.5 arrived with missing safety info and vague transparency details.

As one NYU data science professor put it:

Here’s what we’re seeing instead:

• Shortcuts in safety testing

• Regression in model quality

• “Nanny models” overfit to benchmarks that don’t generalize well

Some models are even hallucinating more than their predecessors—so if your chatbot feels dumber lately, you’re not crazy.

Is it burnout? Have we hit the wall? Let’s see what’s slowing the bots down.

The Usual Suspects: Data, Compute, and a Side of Chips

At a technical level, several issues could be contributing to the drag:

• Training data is getting harder to find—especially the kind that’s diverse and legally usable.

• Compute power is expensive and in short supply, even for the big players.

• Chips are still scarce, and Nvidia controls most of the market.

But while real, these issues don’t fully explain the trend. Because something more fundamental has changed.

From PhDs to IPOs

The AI breakthroughs of the past decade were born in open research environments—Meta’s FAIR lab, Google Brain, and OpenAI (back when it was still a full nonprofit). These teams prioritized exploration over deployment. They were optimizing for knowledge, not KPIs.

That era is largely over.

Today, AI has become a business. And businesses don’t optimize for discovery—they optimize for product cycles, market share, and returns.

• OpenAI tried to go fully for-profit

• Google folded Brain into DeepMind

• Meta gutted FAIR in favor of the GenAI division

Regardless of all the different names of the different labs, the key point is this: the places that used to explore AI for discovery are now building products for profit—and that shift changes what gets built, who builds it, and who it’s built for.

GenAI Plateau: Is It Having a Midlife Crisis?

Even if the answer is yes, other areas of AI—like robotics, autonomous systems, or predictive healthcare—are still advancing. But if LLM’s hit a ceiling, the mainstream momentum behind AI could slow down.

The irony here is that the AI revolution wasn’t triggered by a research paper—it was triggered by ChatGPT, a polished product that put generative AI in the hands of everyday users.

These users probably won’t notice if GenAI doesn’t get a dramatic upgrade every month.

But for companies and the investors backing them this shift matters. Billion-dollar bets need billion-dollar returns. And if this really is a plateau, some of that capital may start looking elsewhere.

We poked at the AI bubble to see if it’s full of innovation or just hot air—check it out here.

💡How To AI: Pro-Level Prompts You’ve Probably Never Tried

• Ask for multiple perspectives.

  “Give me three different takes on this topic—a contrarian one, a mainstream one, and a futuristic one.”

• Set constraints or limitations.

  “Explain quantum computing without using any words longer than 8 letters.”

• Prompt it to teach you first.

  “Before writing the summary, explain the topic to me like I’m in 7th grade.”

• Use it to challenge your own thinking.

  “Here’s my argument—now poke holes in it like a debate opponent.”

• Ask for structure, not just output.

  “What’s a logical sequence for presenting my story in this slide presentation?”

• Ask for analogies.

  “Compare AI fine-tuning to training a puppy.”

• Give it your writing sample.

  “Here’s a paragraph I wrote—match the tone and write the next one.”

• Use “you are” prompts to set a role and tone.

  “You are an overly dramatic 19th-century novelist. Now summarize this press release.”

Craving more ways to prompt like a pro? HubSpot has 100 ready to go.

Want to get the most out of ChatGPT?

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

Download the free guide

Heard in the Server Room

AI Can Now Predict Where Proteins Live in Your Cells

Tracking where proteins hang out inside human cells is key to understanding diseases like cancer and Alzheimer’s—but doing it in the lab is a slow, pricey grind. Enter PUPS, a new AI tool from MIT and Harvard that can guess any protein’s location in any cell, even if scientists have never seen it before. It combines protein “language” with computer vision to make highly specific, single-cell predictions—no microscope required. The impact? It could speed up disease research, help discover new treatments faster, and make biology more accessible to scientists everywhere.

Art Of The Deal: US And UAE Announce Major AI Partnership

While touring the UAE, Trump locked in $200B+ in new U.S.-UAE deals, headlined by a splashy AI partnership aiming to crank up the Gulf state's tech game. The deal includes a plan for the UAE to snag 500,000 Nvidia AI chips per year and pour cash into U.S. data centers that rival its own. With the UAE angling to become the third AI superpower (next to the U.S. and China), the move ups the ante in the global AI arms race. To calm D.C.’s nerves, the agreement also includes guardrails to keep U.S. tech from drifting into Beijing’s hands.

Wisconsin County Uses AI To Handle Non-Emergency Calls

La Crosse County, WI just deployed a multilingual AI phone assistant named Ava to handle calls to its non-emergency line—think parking complaints, lost pets, and noisy neighbors. Officials say Ava could take 40–50% of the more than 140,000 annual calls off human dispatchers’ hands, freeing them up to focus on real emergencies. The system doesn’t replace staff—it just handles the boring stuff and hands you off to a human if needed. Bonus: Ava speaks English, Spanish, and French, and could soon go full Duolingo with support for 30+ languages.

Lost and Hound: How AI Is Reuniting Lost Pets With Their Owners

When Michael Bown’s dog Millie bolted in NYC, it had all the makings of a lost-pet horror story. But less than 14 hours later, she was back—thanks to AI.

Enter Petco Love Lost, a free platform that uses AI-powered photo matching to reunite pets with their humans. Millie’s photo and microchip were uploaded, and boom: the system scanned a huge database and flagged a match at a vet across the city in a few minutes, where she’d been taken after getting hit by a car.

The tech behind this AI tool is a computer vision system that analyzes up to 512 data points in each image and compares them to thousands of shelter and user-submitted photos across the country. Users get real-time alerts when there’s a hit—that means no more doom-scrolling lost pet pages.

Bottom line: Petco’s Love Lost AI tool turns a needle-in-a-haystack search into a quick reunion. One upload. One match. One very lucky pup.