Welcome, humans.

Ever wonder if your job as a software engineer is going the way of the slide rule? Matt Welsh (ex-Google, ex-Apple, now startup founder) just gave what might be the most brutally honest presentation about AI and programming we've seen all year. His thesis: we've run a 50-year experiment proving humans are bad at writing software and LLMs are about to change everything.

Spoiler: he thinks there will be fewer programmers making less money in the next 4-15 years and some even setup a reminder in this Reddit thread. But he's also weirdly optimistic about what that means for the other 8 billion people on the planet who can't code.

Here's what happened in AI today:

Google just invested in Sakana AI, co-founded by a transformer co-author who's now "sick of transformers."
Pope Leo XIV warned against "overly affectionate" AI chatbots becoming emotional crutches.
Apple plans to unveil Google Gemini-powered Siri features as soon as next month.
ChatGPT started pulling answers from Elon Musk's controversial Grokipedia encyclopedia.

Advertise in The Neuron here!

Don't forget: Check out our podcast, The Neuron: AI Explained on Spotify, Apple Podcasts, and YouTube — new episodes air every week on Tuesdays after 2pm PST!

Google Just Invested in the Guy Who Helped Invent Transformers… Who Now Says He's "Sick" of Them

DEEP DIVE: Why is Google teaming up with Sakana AI? A few theories…

So Google just announced a strategic partnership with Sakana AI last week, backing the Tokyo-based startup with funding and access to Gemini and Gemma models. Why Sakana? Because Sakana's CTO, Llion Jones, is one of the eight co-authors of the famous "Attention Is All You Need" paper, and he's publicly declared he's "absolutely sick of transformers."

Here's what makes this interesting:

Sakana isn't building bigger transformers. They're exploring fundamentally different approaches inspired by biology:

Continuous Thought Machines: A new architecture that incorporates timing information at the neuron level (something transformers ignore). When it solves mazes, the model's attention literally traces the path, thinking step-by-step like a human.
The AI Scientist: An automated research system that generated a paper that passed peer review at ICLR (the first fully AI-authored paper to do so).
ALE-Agent: Just this month, this coding agent became the first AI to win a major optimization programming contest, beating 804 human participants.

Why Google is hedging:

Google DeepMind CEO Demis Hassabis recently said at Davos that AGI probably needs "one or two more big breakthroughs", specifically around continual learning (current models have "goldfish brains") and better memory. Those are exactly the problems Sakana's research tackles.

BTW: This interview from Alex Kantrowitz is probably the best interview with Google DeepMind CEO Demis Hassabis to come out of Davos we've seen; we highly recommend you watch it if you haven't yet! Demis also pops off a bit at Sam's wish that AGI is already achieved…:

Fun fact: Of the eight transformer co-authors, only two still work at major frontier labs pushing the current paradigm. The rest have scattered: startups, other AI labs, even RNA drug discovery. Jones, however, is the most vocal about moving beyond what they built.

The partnership will focus on three areas: using Google's models to accelerate Sakana's research, providing feedback to improve Google's AI ecosystem, and deploying AI in mission-critical Japanese industries (finance, government) requiring high security and data sovereignty.

So before you get too excited, Google isn't abandoning transformers; Sakana will actually use Gemini extensively. But Hassabis has said language models (like ChatGPT and Gemini) might be "a key component" of AGI rather than "the only component," and he's called for breakthroughs in exactly what Sakana researches: continual learning, efficient memory, long-term planning.

Google is betting big on AI for science, and Sakana's AI Scientist already generates peer-reviewed research for $15 a paper. If the road to a true AI scientist runs through hybrid approaches, Google just bought itself a seat at that table too.

FROM OUR PARTNERS

Agents that don't suck

Are your agents working? Most agents never reach production.

Agent Bricks helps you build high-quality agents grounded in your data. We mean "high-quality" in the practical sense: accurate, reliable and built for your workflows.

Generic benchmarks don't cut it. Agent Bricks measures performance on the tasks that matter to your business.

Evaluate agents automatically, and keep improving accuracy with human feedback. With research-backed techniques for building, evaluating and optimizing, you can turn your business data into production agents faster — with governance built in from day one.

See how Agent Bricks works

Prompt Tip of the Day

Forget scrolling through news or checking your inbox first thing—try this "brain warm-up" prompt instead.

Productivity writer Alex Morgan shared a simple morning ritual: before diving into your day, ask yourself (or your AI chatbot) one open-ended question that invites curiosity rather than demanding answers. It takes under two minutes and helps break autopilot mode before the chaos begins.

Sample prompts to try:

"Give me one small thing to notice today."
"Share one interesting assumption I might hold without realizing."
"Suggest an overlooked detail worth noticing on my commute."
"Offer a single question that could change my usual perspective."

The key is to pause after asking. Let the reflection settle in the background while you brush your teeth or make coffee, often the clearest insights emerge unconsciously as you go about your routine.

Why this works: Traditional morning routines push you toward checklists or deep meditation, but this takes a different approach. Instead of staring at overwhelming inboxes or forcing productivity, you're gently syncing your mental and physical readiness with one manageable point of contemplation.

Our favorite insight: Over time, this habit builds mental agility and makes you less reactive throughout the day. Change the phrasing weekly to keep it fresh, or invite coworkers to share their morning prompts—some teams are making it a Slack ritual before standup meetings.

Want more tips like this? Check out our Prompt Tip of the Day Digest for January.

Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 600K readers here!

*Stop typing long prompts. Dictate full context and paste a clean, structured prompt into ChatGPT or Claude. Start for free today.
Four new audio models to know:
1. Qwen3-TTS runs text-to-speech locally on your PC with "voice design", so you can prompt speech styles like "whispered" or "enthusiastic" to control exactly how voices sound (requires ~7GB VRAM, supports 10 languages); install it via Pinokio.
2. PersonaPlex-7B-v1 lets you interrupt voice conversations mid-sentence and customize voice/personality without fine-tuning (paper, code).
3. Inworld TTS-1.5 generates voice in <250ms, 30% more expressive, 40% fewer errors—half a cent/minute (details); $5-10 per million characters.
4. Chroma 1.0 responds in <150ms with instant voice cloning from seconds of audio—fully open-source (paper, code).
Fun / cool stuff:
1. REK1 is a live humanoid robot fighting tournament where you can watch robots battle or apply to pilot one yourself—tickets from $30 (general admission) to $1,000 (includes piloting).
2. This is a map of NYC, but isometric! (and vibe-coded).
Technical stuff:
1. APEX-Agents benchmarks AI agents on real professional Google Workspace tasks—like building Excel models or writing client emails—where frontier models currently pass 25% on first try (paper, code).
2. Edison Analysis pulls thousands of clinical trials from ClinicalTrials.gov with a simple prompt, then cleans the data, standardizes categories, and outputs downloadable tables—turning days of manual work into minutes.
3. FathomDEM+ gives you globally consistent, near-LiDAR-quality terrain data for accurate flood risk modeling and infrastructure planning (paper).
This one's for the parents:
1. Stanford AI4ALL (hosted by Stanford HAI and Stanford Pre-Collegiate Studies) teaches current 9th graders about AI through hands-on research projects and mentorship from Stanford researchers, with online or residential summer programs available—applications now open!

Around the Horn

Pope Leo XIV warned against "overly affectionate" AI chatbots that become "hidden architects of our emotional states," calling for regulation to prevent serious emotional bonds between humans and AI companions.
OpenAI's Sam Altman announced several Codex-related product launches starting next week, with initial restrictions to prevent misuse for criminal activities like hacking banks.
Apple plans to unveil its Google Gemini-powered Siri features in iOS 26.4 beta as soon as next month, nearly two years after first announcing the updates at WWDC 2024.
ChatGPT started citing Elon Musk's Grokipedia (the conservative-leaning AI encyclopedia) in its answers, raising concerns about content quality after the platform previously claimed pornography contributed to AIDS and offered "ideological justifications" for slavery.
San Diego Comic-Con and the Science Fiction Writers Association both banned AI-generated content from their competitions after initially softer stances sparked immediate backlash from their communities.
A new Gallup poll found 12% of U.S. workers use AI daily on the job (up from 21% using it occasionally in 2023), with the highest adoption among tech workers (60% frequent use), finance (daily document synthesis), and educators (cleaning up parent communications).

FROM OUR PARTNERS

Mozilla Pioneers: Mozilla is selecting a small group of product builders for an intensive, paid residency focused on AI and the web. Ship fast, work directly with Mozilla's New Products Team, and explore ideas that could become real products in the market. Applications open January 26-February 16. Apply now!