In partnership with
☀️ TRENDING AI NEWS 🚨 Midterms: 76% of voters polled by NBC News said they oppose data centers in their communities. 🤖 Image Model: OpenAI's GPT-Image-2.6 Flare generates images in as little as 6 seconds and adds GIF output. 🛠️ Inference: Together Compiler beat the best InferenceX configurations by up to 2.6x tokens per second. 🏢 Agents: Uber says only 4 rides last quarter were booked by third-party AI agents. |
Out of roughly 320 million rides and deliveries, four were arranged by an AI agent. Hold onto that number, because today's issue is really about the gap between the AI industry's confidence and what real people actually do, vote for, and tolerate.
| 🛠️ Tool of the day | |
Descript - an AI video and audio editor where you cut and rearrange clips by editing the transcript text. It helps creators and teams edit recordings fast, and it has a free plan.
| ⚡ Quick hits | |
AMD reported record Data Center revenue of $8.5 billion, up 97% year over year, in its third-quarter results. Read more →
Anthropic says it is building an open-source system, called FlashProve, that can prove which AI model produced a given output. Read more →
| 🤓 AI Trivia | |
Today's trivia: what does the acronym GIF actually stand for?
The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇
| 🗳️ Data Centers Lost the Midterms | |
Opposition crossed party lines on Tuesday
The AI industry's biggest physical bet just took a public beating. According to NBC News polling on Election Day, 76% of voters opposed data centers in their communities. Candidates who favor moratoriums and ballot measures that add friction to new builds won across the country, and The Verge calls it a stinging blow to a central pillar of the industry.
Texas is the sharpest example. It is the fastest-growing data center market in the US and ground zero for Big Tech's largest builds, yet Democrats won the Senate seat and the governor's race there. MIT Technology Review asks whether that position is now in jeopardy, while Wired reports that data center boosters are already looking to low Earth orbit for comfort. Meanwhile, the pro-AI super PAC network Leading the Future, with a $100 million war chest, barely moved the needle.
The mood is spreading beyond the US: activists in 15 European countries are staging two days of "Stop Dirty Data Centres" protests that began today, with more than 100 groups calling for a moratorium on large projects. Read the datacenter backlash coverage and the Read the full story →
The bottom line
Power, water, and utility bills turned data centers into a kitchen-table issue, and politicians of both parties noticed. Expect slower permits, more local fights, and a louder push for efficiency gains that need fewer new buildings.
Frontier models perform better in Attio’s agent harness.
We gave GPT-6 Astra the same job twice: analyze 100 sales calls and draft the follow-ups, once in Ask Attio and once in ChatGPT.
Ask Attio finished in 4 minutes 49 seconds. ChatGPT took 26 minutes 11 seconds, about 5.4 times as long.
The most ambitious GTM work requires more than frontier intelligence. It requires architecture that can put that intelligence to work at scale.
| 🎨 ChatGPT Now Makes Animated GIFs | |
Images 2.6 adds speed, SVG and markup edits
OpenAI released Images 2.6 along with two new API models, GPT-Image-2.6 Flare and Sunburst. Flare generates images in as little as 6 seconds, and both models add GIF output, SVG export, markup edits and a new complexity parameter.
In ChatGPT, you can now generate a GIF from a prompt or from an uploaded photo, adjust the animation speed, edit individual frames and share the result. Before this, the chatbot could not natively produce a GIF in a single step. SVG export is the sleeper feature here: vector output means logos and icons that scale without going blurry. If you need a quick site to show off your new graphics, our sponsor 60sec.site is an AI website builder that gets you live in minutes.
The bottom line
This is a practical upgrade, not a flashy one. Faster generation plus vector export makes the API far more useful for real design workflows, and it keeps pressure on every other image model.
| ⚡ Together Compiler Squeezes 2.6x More From the Same GPUs | |
Agents keep re-optimizing the forward pass
Together AI released Together Compiler, an agent-driven loop that continually re-optimizes the inference forward pass as traffic patterns shift. In its tests it beat the best InferenceX configurations by up to 2.6x tokens per second, doing so in under 24 hours per model.
The idea is that a static, hand-tuned serving setup goes stale the moment your workload changes. Letting agents keep recompiling means the system adapts to real traffic rather than a benchmark. That pairs nicely with NVIDIA's separate write-up on stacking three optimizations to lift mixture-of-experts throughput 42.5% on a single DGX Spark. Cheaper inference is also the quietest answer to the data center backlash above. See more on AI infrastructure.
The bottom line
If software can deliver multiples of speed on existing hardware, the economics of serving models shift without pouring more concrete. Treat vendor numbers as best-case, but the direction is clearly right.
| 🚗 Uber: Four Rides From AI Agents | |
The agentic future has a reality check. During its Q3 earnings call, Uber disclosed that only four rides were arranged by third-party AI agents last quarter. Uber Eats did a bit better, with around 500 orders from agents during the week of Halloween, out of 320 million total rides and deliveries that week.
The number cuts both ways. It suggests the fear that agents will intermediate Uber out of existence is premature, but it also shows how far real-world AI agents still are from everyday habits. Readers who caught our recent look at the agentic web will see the same lesson: capability is arriving much faster than adoption.
| 🛠️ Open Models and Developer Moves | |
Small and mid-size models keep punching up
Two open releases stand out. webAI's TwIL-8B, an 8.8B reasoning model post-trained from IBM Granite 4.2 8B, scores 98.2 on MATH-500 and 81.6 on LiveCodeBench, though it trails similar-size models on knowledge benchmarks. Ricursive's Ripple-65B Preview, a 65B dense open-weight model under Apache 2.0, posts 80.5 on Terminal-Bench 2.1, ahead of US open models up to 15x its size.
On the tooling side, Meta acquired Conductor, a startup that lets engineers deploy teams of agents working on different parts of a project at once, a clear bet on parallel AI coding. And Hugging Face's CEO told Wired his P(doom) is close to zero, arguing most risk comes from a few players building behind closed doors. Explore more in our open source AI coverage.
The bottom line
Strong reasoning in an 8B model you can run yourself, plus a permissively licensed 65B, means the open-weight ecosystem keeps closing the gap. Benchmarks are not the whole story, so test on your own tasks before switching.
| 🌎 Trivia Reveal | |
The answer is Graphics Interchange Format! CompuServe introduced it back in 1987, and now ChatGPT can make one from a single prompt.
| 💬 Quick Question | |
Would you be fine with a new data center being built in your town, or would you vote against it? Hit reply and tell me where you land - I read every response!
That's it for today. See you tomorrow with more from dailyinference.com, your daily AI newsletter.
| 🎁 Share Daily Inference | |
Share your link with friends. One referral wins the AI Tools Starter Kit. Five referrals win the AI Insider Briefing.