AI people speak a different language. Here’s a translation of 7 words you saw this week 🧵
1/ LLM: a program that read a huge chunk of the internet and learned to predict the next word. ChatGPT, Claude and Gemini are all LLMs.
2/ Agent: an AI that doesn’t just answer, it does things. Clicks, books, sends, buys. Meta’s Muse is an agent.
3/ Hallucination: when AI confidently makes something up. It’s not lying. It just doesn’t know that it doesn’t know.
4/ Token: a chunk of a word. AI reads, thinks and charges you in tokens. Roughly 100 tokens ≈ 75 words.
5/ Context window: AI’s short-term memory. How much of your conversation it can “see” at once. Too long a chat, and it starts forgetting the beginning.
6/ Jailbreak: tricking an AI into ignoring its own safety rules. This week AI labs made a scale to rate how dangerous each trick is.
7/ Compute: the raw computer power AI runs on. It’s why everyone is building data centers, and why Musk wants to put them in space.
That’s AI dictionary #1. I translate AI into human language every week. Follow @Gooslyy so you’re never lost in an AI conversation again.
🚨Sam Altman says there are two ways AI "goes really wrong."
"One is that there's like a loss of control to AI," the OpenAI CEO said, where "we can't get it back."
"The other is like one person or company or country gets way too powerful relative to everybody else."
Dario Amodei: A lot of AI generated videos and images are fake, addictive, and just slop. Anthropic isn't interested in competing in that market.
“Google and OpenAI are fighting it out in consumer. It's existential to both of them.”
“It's existential to OpenAI because that's their whole business. Existential to Google because they have the search business.”
“For enterprise business, there's not really a demand for cats riding donkeys or whatever consumer video people want.”
“If we ever need it, we can just contract a model from someone else.”
“A lot of short-form video is fake, a lot of it's pretty addictive, a lot of it's slop.”
“It's not a part of the market that I'm tripping over myself to get involved in.”
AI companies just admitted their chatbots can be tricked into helping hackers.
Their fix? A scale that ranks how dangerous each trick is.
In human language: the locks aren’t perfect, so now they’re rating the burglars.
We partnered with @fmf_org and its member firms on a new framework to help identify when bypassing an AI system's safeguards could enable serious cyber harm, and prioritize stronger protections.
"...This framework is an important step toward a common methodology for evaluating and comparing the severity of cyber jailbreaks, including the capabilities they unlock and how readily they can be used. Shared approaches like this can help the AI and security communities build a clearer evidence base and prioritize improvements to safeguards." - OpenAI's Sasha Baker, Head of National Security Policy
https://t.co/epOq98YPau
In human language: investors got scared AI is a bubble and wiped $2.5 trillion off the market in one day.
OpenAI’s answer: our revenue grew 70% in a quarter, companies are paying, so we keep building.
Wall Street is worried about the bill. OpenAI is pointing at the customers.
OPENAI'S CFO SAYS IT WILL KEEP BUILDING DATA CENTERS AFTER THURSDAY'S $2.5 TRILLION SELLOFF
Sarah Friar told CNBC that OpenAI's business grew +70% quarter over quarter in Q3, and that enterprise demand doubled over the same period, per Reuters.
"It is saying people are actually using AI and getting value from it."
On the midterms:
"When I think about the midterms, I need to keep thinking about how do we keep investing to meet customer demand."
On communities:
"We are going to keep building out those relationships with communities so we can get the compute we need."
OpenAI said it will stop using non-disclosure agreements with public officials when negotiating future sites, and will give residents access to construction plans.
Friar also said OpenAI is raising fresh capital, pushing back on suggestions it had scaled back its fundraising plans.
Google $GOOGL said Thursday it will invest $17 billion to build its first data center in Wyoming.
In human language: Claude can now look at your bank accounts and explain your spending in plain words.
It can only read. It can’t move, spend, or invest a single cent.
Think of a friend who can see your statements, not one who holds your card.
Introducing Personal Finance in Claude, in beta in the US.
Connect the accounts you choose, and Claude answers your money questions in everyday language, from your own balances and transactions. Access is read-only, and Claude can't move, manage, or invest your money.
In human language: ChatGPT is an AI you talk to. Muse is an AI that does stuff for you — books trips, handles email, buys things.
And Meta can push it to billions of people already on Facebook and WhatsApp.
Downloads ≠ daily use. But distribution is a cheat code.
Muse from Meta, US App Store downloads last week:
Muse 942K
ChatGPT 556K
Gemini 341K
Grok 239K
1.7x ChatGPT, eight weeks after launch. And in Japan it's #1 in Productivity, ahead of ChatGPT $META