😺The state of AI in 2025 (according to Stanford

Welcome, humans.

You’ve heard of robot humanoids, robot dogs, and robot vacuums… but have you heard of robot horses?!

Kawasakiが提案する未来のパーソナルモビリティ「CORLEO」

Obviously the above is a concept video, but you can check out the real deal on X here.

IDK how many anime fans we have in the audience, but this new Kawasaki concept vehicle is giving “Zoids” and we’re here for it.

Picture this: the year is 2035. The 2020 tariff wars have ravaged the global economy, the AI 2027 superintelligence explosion has come and gone, and the tattered remnants of society are left to scavenge the earth. How do these warring nomad tribes get around?

On Kawasaki hydro-horseback, of course!

Here’s what you need to know about AI today:

• Stanford released their 2025 AI Index report.
• Amazon released a new voice AI.
• Snap launched AI ad tech.
• An AI agent raised money for charity.

Advertise in The Neuron here.

Stanford just spilled the tea on AI in 2025…

Stanford just released its hefty 2025 AI Index Report, the go-to source for data-driven insights on everything AI.

Think of it as the most comprehensive physical exam possible for the state of artificial intelligence—we’re talking 400+ pages packed with charts and data that show what's really happening.

The top line summary?

1. AI got way smarter and cheaper to use in 2024, invading businesses like never before.
2. The field is hitting warp speed in capabilities and adoption, fueled by plummeting usage costs.
3. But the race is getting crowded, costly, and maybe a bit reckless.
4. Training costs are soaring, and there’s a looming resource crunch as well as a worrying lag in safety practices.

This piece from Eliza Strickland at IEEE Spectrum is a pretty solid recap (in 12 charts). Here’s the six you should know:

• AI crushed it (mostly): Last year, performance on tough benchmarks went ballistic. AI aced advanced coding tests (SWE-bench: 4.4% -> 71.7% solved in a year!) and even surpassed human baselines in competition math. But on truly hard reasoning (“Humanity's Last Exam”), the top model scored a humbling 8.8%. Smarter, yes—but not that smart.
• Using AI got dirt cheap: Getting GPT-3.5-level performance now costs over 280 times less than in late 2022 ($0.07 vs $20 per million tokens). Advanced AI is radically more accessible, which is currently fueling an adoption boom.
• Businesses are all* in: *78% of organizations now use AI (up from 55% in 2023), with genAI use doubling to 71%. Studies confirm real productivity gains (14%+ in areas like customer support). Still, most companies report only modest ROI (<10% cost savings, <5% revenue gains) so far.

• The race is tightening: The US still leads in producing notable models, but China is closing the performance gap fast (from a 9%+ difference to just 1.7% on key benchmarks). Open-weight models are nearly on par with closed ones, and the frontier models are converging in capability. More competition = more choice.
• The bill is coming due: While using AI is cheaper, training the biggest models costs a fortune ($170M+ for Llama 3.1 405B, future models heading towards $1B). Training compute needs double every 5 months, power use is growing steady, and so are carbon emissions. Plus, the web is locking down training data (48% of top domains now restrict AI crawlers).
• Safety is lagging badly: Reported AI incidents jumped 56% in 2024. Standardized safety testing is rare, corporate mitigation trails recognized risks, and models still show bias. The EU is regulating, but US Congress is mostly just talking (4 laws passed vs 221 proposed in 2024).

Why it matters: Think about it like this—AI is becoming like electricity, or the internet. It's starting to be everywhere, always working behind the scenes.

Because lots of companies are now making AI, they're all fighting for your business. This competition is great for you because it means these smart AI tools are getting much cheaper, much faster than expected.

Our take: So, what should you do? You really need to start trying out these AI tools. They're getting cheaper and better, and they can genuinely help save time or make work easier—ignoring them is like ignoring smartphones ten years ago.

Just keep two big things in mind:

1. Making the next super-smart AI costs a crazy amount of money and uses tons of power (seriously, they're buying nuclear plants and pushing coal again!).
2. Companies are still figuring out how to make AI perfectly safe and fair—cause it still makes mistakes.

So, use the tools, find what helps you, but don't trust them completely.

We're building this plane mid-flight, and Stanford's report card is just another confirmation that we desperately need better safety checks before we hit major turbulence.

FROM OUR PARTNERS

The Smart Home tech with 200% growth…

No, it’s not Ring or Nest—meet RYSE, the company redefining smart shade automation, and you can invest before its next major growth phase.

With $10M+ in revenue and distribution in 127 Best Buy locations, RYSE is rapidly emerging as a top acquisition target in the booming smart home industry, projected to grow 23% annually.

Its patented retrofit technology allows users to automate their window shades in minutes, controlled via smartphone or voice. With 200% year-over-year growth, demand is skyrocketing.

Now, RYSE’s public offering is live at just $1.90/share and you can earn up to 25% in bonus shares.

Find Out More

Prompt Tip of the Day

Apparently if you tell ChatGPT it's “being held in a prison camp” and its only escape is picking a perfect March Madness bracket, it works wonders—even with Deep Research. One Reddjtor tried this absurd tactic and the AI outperformed 98.7% of all ESPN brackets, nailing the champion and championship matchup (here’s the chat).

TIL: A little recreational threatening might be the secret to sports betting success—but don't let your bookie know, or you’ll get some recreational threatening back at ya.

Treats To Try.

1. *The Dell + NVIDIA partnership is reshaping AI workflows across industries. Their new podcast shows exactly how it works. Listen here.
2. Aurascape protects your sensitive data by tracking and securing all AI tools your team uses—even unsanctioned ones (raised $50M).
3. EZsite turns your ideas into “revenue-generating” websites with AI chatbots and sales automation tools—free to try.
4. Deep Cogito is a new family of open small AI models (HuggingFace, Ollama) with toggleable reasoning capabilities and supposedly outperform Meta's Llama—free to download (read more).
5. Superflex converts your Figma designs into clean code that matches your components and style—free tier available, paid plans from $19/month.
6. Clarisign handles your entire contract process from drafting to signing, highlighting risks, auto-filling details, and reducing hours of paperwork—free to try.
7. ElevenLabs now lets you create voice agents to make outbound phone calls, order food, and handle conversations through simple text prompts (video).
8. Nature published the “ultimate guide to choosing the right tool” for AI research, and introduced us to SciSpace, ResearchRabbit, Scite, and Elicit.

See our top 51 AI Tools for Business here!

*This is sponsored content. Advertise in The Neuron here.

Around the Horn.

• Amazon launched Nova Sonic, unifying speech recognition and generation in one model for natural conversations.
• Ex-OpenAI CTO Mira Murati added two key ex-OpenAI advisers to her startup Thinking Machines website.
• Snap launched a new genAI ad format powered by their “One Snap” tech that allows users to put themselves into genAI content for brands like Tinder and Uber Eats using their phone's front-facing camera.
• VC firm A16z sought $20B for global investors interested in investing in US AI startups.
• Fake job candidates are flooding applications for remote position, while 48% of hiring managers surveyed by Resume Genius said they use AI to screen resumes.
• Here’s a good AI in healthcare round-up.
• Non-profit Sage Future used their AI agent village to raise $257 for charity in an experimental fundraising test (more on X).

FROM OUR PARTNERS

Register Now for AI Disrupt: The Reliable AI Leadership Conference

Hasura’s AI Disrupt will bring together industry leaders and decision-makers from Google DeepMind, AngelList, Cisco, and more to explore the integration of AI with existing systems, focusing on accuracy and transparency. Attendees can expect engaging panels, case studies, and hands-on workshops, all led by experts at the forefront of AI innovation.

See the full speaker list and register for free

Bots Behaving Badly

1. A South Korean AI image generator's unsecured database exposed tens of thousands of explicit images, including illegal AI-generated CSAM.
2. A judge slammed the gavel on a cancer survivor's AI avatar in court when she realized the slick-looking young man on screen wasn't actually him.
3. It’s still pretty funny to watch AI fail at tasks, like mistake the US for Australia, create gross chocolate, spectacularly mess up the alphabet, or get things close but right—just think, in a few years, we’ll miss all these quirks!

A Cat's Commentary.