it's hilarious how many have taken the bitter lesson to mean just scaling up transformers. it's about scaling, but innovations in algorithmic primitives can lead to much more efficient scaling
AI is using more and more of the world’s energy. But how much economic value are we getting from AI?
We’re introducing Joules-to-GDP (J2G), a framework that measures how well AI turns energy into economic value and shows which tasks, models, and hardware give the best return.
Across 44 occupations in finance, engineering, law, administrative work, and more in @OpenAI's GDPval, we arrive at three findings:
1) The tasks with the highest economic value are not the tasks with the highest economic value per joule. For example, CAD design earns ~10% more per task than building websites, but the latter earns ~27% more value per joule.
2) The most powerful model isn’t always the best use of energy. Gemma 4 31B delivers 56% of GLM 5.2’s net value with 23% of the energy, for 2.5x the value per joule.
3) More chips don’t guarantee proportionate returns. Moving the same model from 1 to 8 NVIDIA H100s uses 3.2x the energy but cuts completion time by only 22%, with about the same value.
With @SnorkelAI, we’re also releasing GDPWorks, a benchmark for economically valuable tasks across 36 additional occupations.
Joint work co-led by @amirzeinaalii and @Avanika15, with @Mersad_Abbasi, @Jon_Hartley_, John Hennessy, @HazyResearch, and @Azaliamirh.
More details below 🧵
I've always believed that intelligence should come from simple, general building blocks that learn their own abstractions directly from raw data; in other words, the bitter lesson. I think we're finally seeing a real step towards AI that can learn from all data.
Language modeling on raw bytes is a microcosm of this ideal, and to me it's a central problem in AI: if we can't build LLMs that learn end-to-end from bytes, how will we handle the rest of the world's data that we understand even less?
Last year we introduced hierarchical networks (H-Nets) as a first step. Our new version finds up to 3-5x efficiency gains in controlled scaling law experiments, and at larger scale matches an open-source model trained on 11T tokens with ~10x less data and compute. The posts below walk through what we found and why we think it works.
to prepare for an upcoming release (👀) we dug into the pretokenizer a little more. are people aware that hand-crafted regexes are part of every frontier LLM and doing some pretty heavy lifting?
some quick examples 🧵
Introducing Multilingual Voices
We partnered with Baklavastory, our neighbor in the Mission, to show what it sounds like when a business's character and warmth stay consistent, no matter who calls or what language they speak.
It's easy to translate speech, but much harder to preserve a single identity across languages. Every language has different rhythms, tones, and emphasis. When you change the language, identity tends to get lost in that shift, and your voice ends up sounding like someone else.
With Multilingual Voices, pick a voice and keep one brand identity in every market you serve: cartesia.ai/voices
Agentic AI will transform every enterprise–but only if agents are trusted experts.
The key: Evaluation & tuning on specialized, expert data.
I’m excited to announce two new products to support this–@SnorkelAI Evaluate & Expert Data-as-a-Service–along w/ our $100M Series D!
---
Snorkel Evaluate is our new data-centric agentic AI evaluation platform for specialized, mission-critical enterprise settings where vibe checks and out-of-the-box metrics driven by simple LLM prompts are not enough.
Snorkel Expert Data-as-a-Service is our white glove service for expert-level AI datasets, powering frontier LLM developers in areas like expert knowledge, reasoning, agentic action and tool use, and more!
Both built on top of @SnorkelAI’s Data Development Platform, using our programmatic technology to drive higher-quality expert data, faster– for getting specialized AI to real production value.
If you’re building enterprise AI and want to partner around the key ingredient in AI today–the data–book a demo and let's talk! snorkel.ai/demo/
Finally, see thread for details on 🧵👇
- 📽️ A walkthrough of Snorkel Evaluate and Expert Data-as-a-Service on an agentic AI enterprise task
- 📅 An upcoming event on Enterprise Agentic AI with innovators from @Accenture @BNY @Comcast@Stanford@QBE & others
- 📊 An upcoming series of benchmark datasets and model artifact releases
👀 Want early access to the full agentic AI dataset? Retweet this post and we'll send you the link!
🚧 Construction logistics isn’t known for cutting-edge software—but @isheepdog is changing that.
By embedding Numbers Station directly into their product, brokers can instantly query their data using plain English and get real-time insights without waiting on a report or navigating a dashboard.
Our agents eliminated the bottlenecks of manual analysis, and gave their brokers something rare in construction logistics:
Speed, visibility, and confidence.
It’s a powerful example of what happens when traditional industries embrace modern data agents. 🚀
March Madness just tipped off—so we had a little fun and asked our agents to make sense of the data behind the madness. 🏀
We uploaded @EvanMiya’s advanced metrics (O-rating, D-rating, true tempo, 10-0s) into NS, and let our agents surface insights to help us make smarter picks.
In just seconds, we:
📂 Uploaded @EvanMiya’s CSV
📓 Asked in plain English
⚡ Got bracket-ready insights instantly
Use these insights to shape your bracket—and this time, know your picks are backed by advanced agents.
Because whether it’s March Madness or market strategy, smart decisions start with better analysis.
Need help with your bracket?
Grab the dataset in the thread ↓
🚴🏻♂️ I love Strava. It keeps me motivated, helps me track my runs, and gives me a sense of progress. Being in the data industry, I wanted to see my metrics.
But after looking into the in-app analytics, I realized I had more questions than Strava could answer.
Strava has some summary stats, but I wanted to know more—so I exported my Strava data and uploaded it into Numbers Station.
In just a few minutes, I used plain English to analyze my training habits, create charts on-the-fly, and found cuts of data I hadn't found on Strava.
Our agents got to work, and I learned that:
🏃🏻 I ran 2,881 miles; 119 of those on a trail
🚴🏻♂️ My go-to exercise is running and the occasional riding
🗓️ I tend to exercise on Wednesdays
- David Kucher, Principal Engineer
🧵 Read about what other insights I found:
🚀 OpenAI’s Response APIs + Numbers Station Agents
Numbers Station's extensible agent architecture means we can move fast. By integrating @OpenAI's newly released Response APIs, we show how we can seamlessly marry document agents with our structured data agents.
📂 Leverage documents to enhance answers with embedded insights
🔎 Agents use RAG to find relevant data
🔗 Bridge structured + unstructured data
Built for adaptability. Built for better insights.⚡
We're launching new tools to help developers build reliable and powerful AI agents. 🤖🔧
Timestamps:
01:54 Web search
02:41 File search
03:22 Computer use
04:07 Responses API
10:17 Agents SDK
Excited to partner with @NumbersStnAI to launch Model Switch! Powered by models hosted on Together AI, we're giving data teams the freedom to choose the most efficient open-source models for their AI-driven analytics.
2025 is the year of agents, and we're excited to see Numbers Station bring agents to structured data. 🚀📊
🚀 More Flexibility, More Choice: Introducing Model Switch with @togethercompute
We’re excited to partner with Together AI to bring more powerful models to Numbers Station.
🧠 Use your preferred model for chat and vision calls—like Llama 70B, Qwen 72B, and Deepseek (more on the way!)
🚀 Reduce costs & get faster responses—pick the best open-source model for your use case
🔓 No model lock-in—full flexibility as models evolve and emerge
A sneak peek at Numbers Station Workflows.
🚀 Getting insights is just the start—true impact comes from automation.
Instead of answering the same questions over and over, Workflows automates analyses without you ever having to ask.
Setting up a Workflow is simple: 🧵
The future isn’t about adding another BI tool to "wow" executives with a few new insights.
It’s about unifying what you already know and making it accessible and actionable at scale.
Only after unlocking 90% of already found insights does the last 10% of new insights matter.