Tomasz Tunguz - A Series of Unfortunate Decisions
Tomasz TunguzVenture Capitalist If you were forwarded this newsletter, and you'd like to receive it in the future, subscribe here. A Series of Unfortunate Decisions
When a person asks a question of an LLM, the LLM responds. But there’s a good chance of an some error in the answer. Depending on the model or the question, it could be a 10% chance or 20% or much higher. The inaccuracy could be a hallucination (a fabricated answer) or a wrong answer or a partially correct answer. So a person can enter in many different types of questions & receive many different types of answers, some of which are correct & some of which are not. In this chart, the arrow out of the LLM represents a correct answer. Askew arrows represent errors. Today, when we use LLMs, most of the time a human checks the output after every step. But startups are pushing the limits of these models by asking them to chain work. Imagine I ask an LLM-chain to make a presentation about the best cars to buy for a family of 5 people. First, I ask for a list of those cars, then I ask for a slide on the cost, another on fuel economy, yet another on color selection. The AI must plan what to do at each step. It starts with finding the car names. Then it searches the web, or its memory, for the data necessary, then it creates each slide. As AI chains these calls together the universe of potential outcomes explodes. If at the first step, the LLM errs : it finds 4 cars that exist, 1 car that is hallucinated, & a boat, then the remaining effort is wasted. The error compounds from the first step & the deck is useless. As we build more complex workloads, managing errors will become a critical part of building products. Design patterns for this are early. I imagine it this way : At the end of every step, another model validates the output of the AI. Perhaps this is a classical ML classifier that checks the output of the LLM. It could also be an adversarial network (a GAN) that tries to find errors in the output. The effectiveness of the overall chained AI system will be dependent on minimizing the error rate at each step. Otherwise, AI systems will make a series of unfortunate decisions & its work won’t be very useful. |
Older messages
The Fastest Growing Category of Venture Investment in 2024
Thursday, May 9, 2024
Tomasz Tunguz Venture Capitalist If you were forwarded this newsletter, and you'd like to receive it in the future, subscribe here. The Fastest Growing Category of Venture Investment in 2024 The
The Capex Conquest in the Cloud
Wednesday, May 1, 2024
Tomasz Tunguz Venture Capitalist If you were forwarded this newsletter, and you'd like to receive it in the future, subscribe here. The Capex Conquest in the Cloud Amazon announced their
Massive Acquisitions in Software Startups
Monday, April 29, 2024
Tomasz Tunguz Venture Capitalist If you were forwarded this newsletter, and you'd like to receive it in the future, subscribe here. Massive Acquisitions in Software Startups What drives the
When 1% Market Share Shifts Represent $5b of Market Cap
Friday, April 26, 2024
Tomasz Tunguz Venture Capitalist If you were forwarded this newsletter, and you'd like to receive it in the future, subscribe here. When 1% Market Share Shifts Represent $5b of Market Cap If it
Partnering with Dropzone: Automating Security Operations with AI
Thursday, April 25, 2024
Tomasz Tunguz Venture Capitalist If you were forwarded this newsletter, and you'd like to receive it in the future, subscribe here. Partnering with Dropzone: Automating Security Operations with
You Might Also Like
Chatting With Her - The ChatGPT App on Mac
Monday, May 20, 2024
Tomasz Tunguz Venture Capitalist If you were forwarded this newsletter, and you'd like to receive it in the future, subscribe here. Chatting With Her - The ChatGPT App on Mac For the past few
📂 How to test new pricing (and migrate legacy pricing)
Monday, May 20, 2024
Today's newsletter is proudly supported by xFusion 🎉 A few years ago I referred one of my consulting clients, SavvyCal, to xFusion to get help with customer support. And after seeing
When to Raise a Series A? 16 Silicon Valley Startups Raised $622.9M Last Week
Monday, May 20, 2024
💰 Sigma Computing Raised $200M Series D 💰 Top Universities For Funded Founders 🔢 Social Capital Fund Return (2011-2023) 👑 Palmer Luckey Wants to Be Silicon Valley's War King ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏
🦄 3D bioprinting for customizable tissue
Monday, May 20, 2024
FoldInk is building material, called bioinks, which are used to print real tissue.
NEW: state of AI
Monday, May 20, 2024
funding and deal size jumps, M&A exits drop, and six new unicorns join the club The State of AI Take a deeper look Hi there, Our newest State of AI Report is out. Spoiler: AI funding hit a 4-
How the customized learning pathway works
Monday, May 20, 2024
This is life-changing Hi , Many are asking how the customized learning pathway inside foundr+ works so I wanted to share. Here's how it works: Initial Assessment: When you first log in, you'll
April 2024 updates, new product!
Monday, May 20, 2024
Traveled to Bali and Sydney, some updates on Typing Mind, and a new product. ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏ ͏
Navigating hurdles as a migrant founder
Monday, May 20, 2024
Plus: Europe's first Black solo GP and the plant-based meat startup aiming for profitability View in browser Sponsor Card - Flagship (44)-2 Good morning there, First-generation migrants are a tough
The Untold Stories of Y Combinator
Sunday, May 19, 2024
Garry Tan interviews Jessica Livingston, co-founder of YC | This Week at YC May 19th, 2024 The intro bit this week is going to be nice and quick because I'm just emerging from the gnarliest cold I
Initiator Creator - Issue 142
Sunday, May 19, 2024
Initiator Creator - Issue #142 - ( Read in browser ) By Saurabh Y. // 19 May 2024 Presented by DesignThingy This Week's Notes: Make GTM Strategy for Yourself GTM or Go-to-market strategy is