GPT-5 Has a Bumpy Takeoff, Here’s What Went Wrong

FEATURED INSIGHT💡

GPT-5 Rollout: From “Way Dumber” to Damage Control

The much-hyped GPT-5 didn’t exactly stick the landing. Users reported sluggish rate limits, missing features, and (according to Sam Altman) a broken auto-switcher that made the model “way dumber” on launch day.

OpenAI scrambled to smooth things out. They:

• Doubled rate limits for Plus users

• Brought back GPT-4o as an option for those who prefer it

• Added a new UI indicator to show which model is answering

• Added easier manual “thinking” triggers for complex reasoning

Altman admits the rollout was “a little more bumpy than we hoped,” citing massive demand (API traffic doubled in 24 hours) and ongoing fine-tuning of how GPT-5 chooses when to hand off tasks to other models.

ON THE HORIZON 🌅

Sam Altman on the AI Bubble: “AGI Isn’t the Point”

In a CNBC interview, the usually bullish Altman backed away from the AGI race rhetoric, calling the term “not super useful” and shifting focus toward steady, exponential gains in model capabilities.

The context:

• Apple’s June study found advanced models “collapse” on complex tasks

• Nvidia research suggests smaller, cheaper models can perform competitively with LLM-based agents

• Running giant models is costly — OpenAI expects $20B ARR this year but isn’t yet profitable

Critics warn of dot-com-era parallels, but with U.S. subsidies, streamlined permitting for AI data centers, and federal AI race framing, big AI players may already be too big to fail.

LATEST IMPORTANT NEWS 📰

Perplexity’s $34.5B Chrome Offer

The AI search startup’s unsolicited bid is likely more legal maneuver than genuine acquisition attempt, aiming to show regulators Chrome has willing buyers if ordered to be sold.

Cohere’s $6.8B Valuation

The enterprise LLM specialist closed a $500M round with AMD, Nvidia, and Salesforce participating. Known for security-first AI, it’s pulling heavyweight hires from Meta and Uber.

US Government Eyes Intel Stake

Talks are underway for a government investment to boost U.S. chip manufacturing — just days after Trump called for Intel’s CEO to resign over alleged China ties.

FOR THE TECHNICALLY INCLINED 🛠️

Google’s LangExtract: Structure from Chaos

If you’ve ever tried to wrangle unstructured text into clean, structured data, you know it’s messy work. Google’s new open-source Python library, LangExtract, makes it easier — supporting Gemini, OpenAI, and Ollama models.

Why it matters:

• Handles massive inputs (full books) with multi-pass extraction

• Enforces custom schemas, supports nested attributes

• Offers visual entity review and precise source grounding

• Runs locally or via cloud APIs

From clinical notes to literary analysis, it’s a versatile addition to the AI developer toolkit.

6 free tools to communicate better at work

Smart Brevity is built to fix inbox — and information — overload. Its science-backed methodology can take any communication from confusing to clear.

Unlock our free resources on the communication…

• Method that makes work more efficient

• Tactic that hooks busy readers

• Format that structures sharper updates

Start making every word work harder so your readers don’t have to.

Access the hub

AI TOOL OF THE DAY 🚀

Ztalk.ai — Real-time voice translation for video calls, seamlessly integrating with Zoom, Google Meet, and other platforms. Break language barriers without breaking your flow.