Jargon stinks.  What do the terms open weights, RAG, and agent mean exactly? Here’s a plain English, slightly snarky glossary of befuddling AI terminology with references for further reading.

AI is a broad name for the technology. Machine learning is the part where a system learns from data instead of following rules somebody wrote, a neural network is the structure that does the learning, and deep learning just means a neural network with a lot of layers.

Here’s the nitty-gritty: the terms that get used loosely, and the distinctions the loose usage hides.

1. Model, LLM, frontier model

ChatGPT is the app you open; an LLM, or large language model, is the AI running inside it.

“Frontier” isn’t a technical category at all. It means the handful of biggest and most capable models at any given moment, so the trophy keeps changing hands.

Everyone says “LLM” and hardly anyone could define it on the spot. “Frontier model” is worse. It’s a ranking, announced by the people being ranked.

Further reading: How ChatGPT Works: A Non-Technical Primer (MIT Sloan). Rama Ramakrishnan walks through the predict-the-next-word mechanism everything else is built on.

2. Prompts, tokens, parameters

A prompt is the thought, question, or instructions you provide to the LLM (plus whatever the app added before it without telling you). The LLM takes the prompt and generates words, both in its internal “thinking” process and in the answer it shows you.  

Tokens are (roughly) the words going in and coming out. The model chops your prompt into tokens, then produces more of them as it answers, and they’re what the industry charges by.

Parameters, also called weights, are the numbers inside the model. A frontier model has hundreds of billions of them and the biggest now run to trillions, and nobody can tell you what any single one does.

Parameter counts get quoted like horsepower. The number nobody advertises is how many tokens it takes to answer your question, and that’s the one that shows up on the bill.

Further reading: The only AI glossary you’ll need this year (TechCrunch, July 2026). Its entries on tokens and weights are the clearest short treatment of the building blocks.

3. Pre-training, post-training, fine-tuning

Pre-training is feeding the model most of the internet, so it learns to predict the next word in a sentence. That’s the expensive part, and it produces something that knows a great deal but can’t follow an instruction.

Post-training is where people rank its answers and it learns to give more of what ranked well. Fine-tuning is post-training done by you, to somebody else’s model, on your data.

Pre-training costs hundreds of millions and gets you a model that won’t answer a question well. Post-training is what gets you the product.

Further reading: Illustrating Reinforcement Learning from Human Feedback (RLHF) (Hugging Face, 2022). The clearest walk-through of how ranking a model’s answers becomes a signal for training.

4. Training from scratch vs. distillation

From scratch, you buy (or rent) the computers and do the work to build and train a model. Distillation trains a cheap model on an expensive model’s outputs, so it inherits the behavior without the bill. Distillation is against most AI companies’ terms of service.

OpenAI accused DeepSeek of distilling its models, which is a bold position for a company that trained on the whole internet without asking. Learning from other people’s work is fine right up until the other people are you.

Further reading: OpenAI accuses DeepSeek of “free-riding” on American R&D (Rest of World, February 2026). OpenAI’s memo to Congress, and an analyst’s reply that no model is an island.

5. Training vs. inference

Training is how you build a model. Inference is what happens every time it answers: the model runs and produces a result.

Training is a one-time cost. Inference is a cost you’ll pay forever. Training runs for months and costs hundreds of millions; one inference, meaning one answer, costs a fraction of a cent, and it happens billions of times a day.

Training costs get announced. Inference costs get discovered. Only one of them shows up in a press release.

Further reading: Why AI’s next phase will likely demand more computational power, not less (Deloitte, 2025). Inference reaches about two-thirds of all AI compute in 2026, up from a third in 2023.

6. Open weights, open source, API-only

We typically use LLMs by accessing an app like ChatGPT, Claude, or Gemini. But experts often want the model itself, not just an app wrapped around it. Open weights means that an AI expert can download the model and run it on a server. You don’t get the data or the code that made it.

Open source means data and software that experts can use and modify, which almost no major model offers (AI2’s Olmo is a rare exception).

API-only means you can’t have the model at all. You send your text to the company’s computers, the answer comes back, and you pay for every use, which is also what’s happening when you use ChatGPT or Claude through an ordinary account.

Open weights is how you claim the open-source mantle without giving much away. Open washing, basically.

Further reading: Open-Weight Models Aren’t Enough. We Need Truly Open Source AI Models for Science and Society. (Stanford HAI, August 2026). James Landay’s term for downloadable weights without the data or code is “open distribution.”

7. Context window, memory, RAG

The context window is how much text the model can hold in mind at once, including your question and everything pasted into the conversation.

Memory is a feature that saves facts about you and slips them back into the context window later.

RAG, short for retrieval-augmented generation, searches a document collection and drops the relevant passages into the context window before the model answers.

Nothing in the model remembers you. The app keeps a file on you and pastes it in before every conversation, and that’s a less charming way to describe the same feature.

Further reading: Glossary of Terms: Generative AI Basics (MIT Sloan Teaching & Learning Technologies). Defines context window and RAG in plain language, and is careful to put the model’s “memory” in quotation marks.

8. Chatbot, workflow, agent

A chatbot answers and stops. A workflow runs the steps you defined, in your order. An agent receives a goal instead of steps, and works out for itself what to do, calling out to other software and checking the results until it’s done or stuck.

Ask about a delayed flight and a chatbot quotes you the policy; a workflow uploads the refund form you built; an agent rebooks you.

Useful test: if it decides its own next step, it’s an agent. If you decided the steps, it’s a workflow.

Further reading: Building effective agents (Anthropic, December 2024). The source of the distinction: workflows run predefined code paths, agents direct their own.

9. Hallucination, AI slop, AI cream

A hallucination is a confident falsehood, like a citation to a paper that doesn’t exist. The model isn’t lying; it has no notion of truth to violate. It’s producing text that looks like the right kind of answer.

AI slop is a different failure: accurate, fluent, and worthless. Think of the LinkedIn post that says nothing in 300 fluent words.

AI cream is the third case and the rare one: superb writing authored with the help of AI.

Nobody sets out to make slop. Everyone believes they’re making cream.

Further reading:  2025 Word of the Year: Slop (Merriam-Webster, December 2025). The dictionary definition turns on quantity: low-quality content “produced usually in quantity” by AI.

Why language models hallucinate (OpenAI, September 2025). Argues that hallucinations persist because benchmarks score accuracy alone, so guessing beats admitting ignorance.

10. Alignment, guardrails, censorship

Alignment is the research problem of getting a model to do what people want when nobody’s watching. Guardrails are the rules behind its refusals: “no, I won’t tell you how to make a bio weapon.” Censorship is a guardrail that blocked something you wanted.

The same refusal is “safety” in the press release, “guardrails” in the documentation, and “censorship” on X.

Further reading:  Model Spec (OpenAI, updated December 2025). A published rulebook for what one model will and won’t do, which makes refusals arguable rather than mysterious.

I snuck in one novel term that’s been sorely absent from the field.  Can you tell which one?

Further reading: other glossaries

Five general AI glossaries, listed roughly from most opinionated to most technical.

The only AI glossary you’ll need this year (TechCrunch). About 30 entries, written for readers who follow the industry news. Strongest on distillation and compute.

Artificial intelligence glossary: 60+ terms to know (TechTarget). The broadest of the mainstream lists, and the only one that bothers to define model collapse.

Glossary of Terms: Generative AI Basics (MIT Sloan Teaching & Learning Technologies). Twenty-odd entries aimed at people who use the tools rather than build them.

Glossary of Terms for Artificial Intelligence (Columbia Business School). The shortest and plainest. Useful as a test of which terms are unavoidable.

Machine Learning Glossary (Google for Developers). Hundreds of technical entries, and the only glossary here that defines “AI slop” a few lines away from several hundred pieces of real math.

TikTok is laying off 75 workers in the Seattle region, focused largely on the company’s e-commerce business, according to a notice filed Tuesday with Washington state.

Job titles listed in the notice are almost entirely TikTok Shop and Global E-Commerce roles in Bellevue, Wash., including anti-fraud and governance program managers, seller and creator operations staff, campaign managers, data scientists, and backend and frontend engineers.

It’s part of a steady stream of tech layoffs this year. Zillow cut more than 500 jobs this month, including 91 in Washington state. Microsoft eliminated 605 positions in the state in July as part of a broader reduction of 4,800. Google cut 52 jobs and Salesforce cut 59 locally this month.

The notice Tuesday was filed by TT Commerce & Global Services LLC on TikTok letterhead, and lists two ByteDance employees as contacts. It gives the affected facility as Lincoln Square North at 700 Bellevue Way NE, with a separation date of Oct. 19.

TikTok Shop is the company’s in-app shopping business, which lets brands and creators sell products directly in TikTok videos and livestreams. The company has used the Seattle region as a base for the e-commerce push, expanding its Bellevue offices as it built out the business.

GeekWire has contacted TikTok representatives for comment, and asked for details on the size of the company’s remaining workforce in Bellevue and the Seattle region.

The cuts follow TikTok’s announcement on Aug. 6 that it will close its Nashville office and lay off all 250 workers there, most of them on content moderation teams.

The company last year cut 65 Seattle-area jobs, including 38 at TikTok and 27 at ByteDance.

Jeff Dean has spent decades flexing his brain to solve some of computing’s hardest problems. Turns out the Google legend and UW computer science alum has been pushing his physical limits just as hard.

In an exchange on X this week, the former Google chief scientist — who recently left the tech giant after 27 years to launch the scientific AI startup Discovery Loop — laid out a weekly training routine that looks less like a tech executive’s calendar and more like a triathlete’s.

In addition to running or biking to work three or four directions a week, Dean, 58, says he’s mixing in a couple trips to the gym, a couple yoga sessions, and a couple games of soccer in a 25-and-over league. He also does a longer run or bike ride on the weekend.

We got winded just scrolling through the comments.

Soccer (2 games/wk in my 25-and-over league), gym 2x/wk, yoga 2x/wk, run to work 3-4 directions/wk (8k each way), bike to work 3-4 directions/wk (8k each way), longer run or bike on weekend.

— Jeff Dean (@JeffDean) August 11, 2026

People wanted to know how Dean took care of his knees (squishy shoes), whether he came up with great ideas while exercising (definitely), when he showers (at work), and if he sleeps (usually seven hours).

In GeekWire’s Slack channel this morning, co-founder John Cook, a longtime soccer player, expressed his own surprise.

“He’s playing soccer twice per week in an over-25 league — that is total bad ass,” Cook wrote. “I can barely run in my over-50 league!”

Dean earned his Ph.D. in computer science from the University of Washington in 1996 before joining Google as employee No. 30. He achieved tech-legend status by helping to build the company’s core computing architecture over nearly three decades.

His new venture, Discovery Loop, is an ambitious public benefit corporation aimed at using AI to automate the scientific method — running thousands of parallel experiments to accelerate breakthroughs in fields like drug discovery, materials science, and machine learning.

Dean’s dive into his exercise routine came up because of a slide he shared at KDD 2026, a conference in Jeju, South Korea. The slide was a play-by-play of his final 48 hours at Google (Aug. 5–7), detailing the whirlwind around his departure and startup launch.

Along with breaking the news to Google colleagues and announcing the launch of Discovery Loop, Dean spent last Wednesday responding to hundreds of well wishes — while squeezing in a yoga session and a soccer game. He went to bed at 2:30 a.m. and was up at 6:30 a.m. Thursday to respond to hundreds more messages.

He was technically unemployed for exactly one second at midnight on Thursday.

He probably spent it stretching.

Indeed, it has been quite a week, so I thought I'd share it with the KDD 2026 audience! For my Google colleagues who might have sent me internal chat messages after I closed my laptop for a final time on Thursday afternoon, I apologize for my lack of response!

(Zoomed in) https://t.co/lnBtuqM0q5 pic.twitter.com/VqyLMPBDSV

— Jeff Dean (@JeffDean) August 11, 2026

Speaking in June at the University of Washington Allen School commencement, AI pioneer Jeff Dean told computer science graduates how he “got the itch to join a startup in 1999,” landing at Google when it had a grand total of 20 people above what is now a T-Mobile store in Palo Alto.

Twenty-seven years later, now 58, the UW alum has the itch again.

Google announced Wednesday that Dean, its chief scientist, is leaving with three colleagues to launch Discovery Loop, a startup automating the process of scientific research: proposing experiments, running them, evaluating results, and iterating, thousands of times over.

“Imagine a future where a handful of people can conduct scientific research and engineering tasks much more rapidly, and with higher quality, than massive teams of scientists and engineers do today,” they write on their website.

Discovery Loop is based in Palo Alto, with what it describes as a lean team of its own. Joining Dean are Google senior fellow Sanjay Ghemawat, his collaborator of more than two decades; Google DeepMind research VP Oriol Vinyals; and Google Brain co-founder Quoc Le.

Google is a founding investor and will supply computing power for at least the first year. Discovery Loop is structured as a public benefit corporation, with backing from Radical Ventures and Khosla Ventures (a name Seahawks fans will recognize from the team’s incoming ownership group).

The departures came as part of a broader shakeup announced in a memo from CEO Sundar Pichai. Demis Hassabis is handing off day-to-day leadership of Google DeepMind to become its chair and Alphabet’s chief scientist, while CTO Koray Kavukcuoglu steps up as SVP, overseeing Gemini model development. Alphabet shares fell about 4% after the announcement

“After an incredible 27-year run, Jeff Dean is at a moment where he wants to try something new, and we’re excited to support him in that,” Pichai wrote.

Dean, who earned his UW computer science Ph.D. in 1996, gave no hint of his plans at the Allen School commencement. He told Wired the idea came together only in recent weeks.

But the general theme was there. Listing problems he thought worth solving, he pointed the UW graduates toward “developing tools that accelerate scientific discovery and engineering.”

“One of the beauties of software,” he said during his commencement address, “is that small groups of people can build things that have enormous impact in the world.”

— Teri Hatfield was named chief revenue officer for Iterable, a customer engagement platform. The Seattle-area tech executive most recently served as executive vice president of sales and solutions at Salesforce and CRO of Tableau, which Salesforce acquired in 2019. She joined Tableau in 2012 and spent more than a decade in sales leadership roles at Verizon earlier in her career.

“I’ve come to believe that great customer relationships don’t start with technology. They start with understanding your customers,” Hatfield said on LinkedIn. “Iterable stood out because it’s built around that belief.”

Hatfield will work remotely for Iterable, which is based in San Francisco.

— Seattle-area attorney Ken Miller has joined Arnold & Porter as a partner in the life sciences and technology transactions teams within the law firm’s corporate and finance group.

Miller began his legal career at Perkins Coie, where he spent 16 years focused on tech companies and nonprofit organizations.

In 2015, he moved to the Gates Foundation as associate general counsel, briefly serving as lead counsel for the Gates Medical Research Institute before returning to the foundation as director of legal. In that role, he led work tied to the Global Health Division, the research institute, and business development and licensing, privacy and AI.

— Ben Reed is now chief marketing officer for AIM Intelligent Machines (AIM), a Seattle-area startup developing software that lets bulldozers and excavators operate on their own.

Reed previously ran a firm offering AI marketing and go-to-market advisory services. Other past roles include CMO for Sanctuary AI and a decade at Microsoft, where he departed as head of strategic storytelling for the company’s digital transformation platform.

“I’ve spent much of my career helping frontier technologies become understandable, credible, and consequential, from Microsoft Surface and HoloLens to physical AI, humanoids, and robotics. At AIM, the unusual opportunity is that the technology, customers, deployments, and proof already exist,” Reed said on LinkedIn.

— Aerospace startup Gravitics named Philip Wong chief financial officer. The Marysville, Wash.-based company designs and manufactures modular space infrastructure such as commercial space station modules, cargo-carrying spacecraft and orbital carriers. Its customers include U.S. Space Force and Axiom Space. On Tuesday, Gravitics announced a partnership with Lockheed Martin on a Department of War contract.

Wong joins Gravitics from Viasat, a communications company providing satellite internet services, where he led financial strategy and investment planning. He previously held leadership roles at Pure Storage and Seagate Technology and is based in California.

“Philip has spent 30 years deploying capital across satellite, network, and space infrastructure … He knows how to translate real engineering programs into the financial strategy that supports them,” said Colin Doughan, co-founder and CEO of Gravitics, in a statement.

— Ian Fliflet, former chief growth officer at Seattle online sales platform OfferUp, is sharing his insights on Intro, a service that provides video chats with experts from wide-ranging backgrounds.

— Katy Brown, president of Microsoft’s Americas Markets & Industries organization, was named to Avanade’s board of directors. Brown has been with the tech giant for nearly 30 years and is past president and board chair of the Professional Businesswomen of California. She also serves as executive sponsor of the Bay Area Women at Microsoft group.

— Ben Minicucci, CEO and president of Alaska Air Group, has joined Lyft’s board of directors. Minicucci has spent more than two decades with the airline, which is the parent company of Alaska Airlines, Hawaiian Airlines and Horizon Air.

— Seattle tech leader Ash Wahi was appointed to MoPOP’s board of directors. Wahi is the founder and CEO of Revenaut, a startup building an AI marketing tool. His career includes leadership roles in product and advertising at Microsoft, Snap and Meta.

— TiE Seattle, a nonprofit supporting entrepreneurship, named five new members to its board of directors:

• Joseph Sirosh, CEO of CreatorsAG and former executive at Amazon and Microsoft. He also serves on the board for the biotech company AbSci.
• Monika Panpaliya, partner director of product management at Microsoft and former leader at JPMorgan Chase & Co., Boeing and T-Mobile.
• Vamshi Reddy, CEO of Quadrant Technologies, founder of Seattle Venture Capital, and past leader at Lenora Systems and Microsoft.
• Prasad Anguluri, co-founder of the social media platform Haply, which connects neighbors. Anguluri also serves as a Bothell City Council member, president of ANG Technologies, and co-founder of Innovative Investing Group.
• Sanjay Puri, who was previously a vice president with Icertis and a former leader at Avalara, 9Mile Labs, Edifecs and others.

— JT McCrone, a Fred Hutch Cancer Center genomic epidemiologist, was named leader of Nextstrain, an open-source project that tracks the evolution and spread of viral and bacterial pathogens in real time.

The effort is primarily based in Seattle at Fred Hutch and two institutions in Basel, Switzerland. It has received $1.5 million from the Gates Foundation to support its next phase, which aims to make analyses of viral evolution more accessible and scalable.

— Graham Littlehale is now an investing partner with venture firm Felicis. He previously served as vice president of Point72 Ventures.