Dec 21, 2024 · 26m · latent-space
The State of AI Startups in 2024 [LS Live @ NeurIPS]
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
At Latent Space Live during NeurIPS, Sarah Guo and Pranav Reddy of Conviction present a comprehensive retrospective on the 2024 AI landscape, examining foundational model competition, test-time compute scaling, and the structural advantages agile startups hold over legacy SaaS incumbents.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The hosts as informed peer 0.3 Guest teaching 0.0 Guest disagreement 1.3 The hosts pushing back 0.0
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 16:20 Sarah rejects the GPT wrapper narrative
Sarah forcefully pushes back against the pervasive industry dismissal of application-layer startups as mere 'wrappers', asserting that the odds strongly favor specialized product builders.
Hardest push from the hosts ▶ 0:15 Swyx frames the gap in industry content at NeurIPSSwyx lightly challenges the typical academic orientation of conference gatherings, setting up the industry-focused perspective of the session.
Biggest teaching moment ▶ 3:40 Pranav analyzes the shifting model leaderboard and spending sharePranav educates the room with concrete ELO and Ramp spend metrics showing how the assumption of an unbreakable OpenAI monopoly dissolved across 2024.
The host holds their own ▶ 0:45 Swyx contextualizes Sarah's long investing track recordSwyx demonstrates his deep industry context by citing Sarah's decade-long investment history from Greylock to Conviction.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Welcome and Speaker Introductions | 2 | 0 | 0 | 0 | Swyx opens the session with a friendly, collegial intro highlighting Sarah Guo's long venture track record from Greylock to Conviction and framing the purpose of the industry session at NeurIPS. The dynamic is completely supportive and welcoming. | |
| Foundation Model Competition and Token Deflation | 0 | 0 | 1 | 0 | Pranav Reddy delivers a presentation breakdown comparing 2023 to 2024, noting the competitive closing of the foundation model gap, open source advances, and token cost deflation. As a presentation monologue, host-side metrics are zero. | |
| Emerging AI Modalities and New Scaling Paradigms | 0 | 0 | 1 | 0 | Pranav and Sarah highlight new model modalities including Chai in biology, low-latency voice, SWE-bench performance leaps, and test-time compute scaling. The delivery is informative and structured as an analytical keynote. | |
| AI Startup Funding Realities and Growth Categories | 0 | 0 | 2 | 0 | Pranav and Sarah challenge the conventional narrative that the entire AI sector is in an irrational bubble by separating foundation lab mega-rounds from rational application funding. Sarah also highlights the latent demand unlocked by creative tools like Midjourney. | |
| Dissecting the Value Layer: Startups vs. Incumbents | 0 | 0 | 3 | 0 | Sarah delivers strong contrarian takes against common venture orthodoxies, rejecting the 'GPT wrapper' dismissal, explaining why incumbents lack crucial reasoning-trace data, and showing how seat-based pricing faces an innovator's dilemma. | |
| Software Through Data and Future Startup Opportunities | 0 | 0 | 1 | 0 | Sarah synthesizes the core thesis of 'software through data', urging founders to exploit the speed advantages startups hold when navigating rapid paradigm shifts and unbundling legacy workflows. The segment concludes on an encouraging call to action. |
Statements from this episode (12)
Guo: Emerging VC firms can outmaneuver incumbents with outdated mental models
“Maybe that there'd be an advantage versus the incumbent venture firms in that when the floor is lava, the dynamics of the markets change, the types of products and founders that you back change it's a lot for existing firms to ingest. And a lot of their mental…”
Reddy: OpenAI's share of enterprise LLM spend dropped from 90% to 60%
“And the opening I spend at the beginning, at the end of last year in November of 23 was close to 90% of total volume. And today, less than a year later, it's closer to 60% of total volume.”
Reddy: Flagship OpenAI API costs fell 80% to 85% in roughly 18 months
“This is a graph of flagship OpenAI model costs, where the cost of the API has come down roughly 80, 85%, and call it the last year, year and a half which is pretty remarkable.”
Reddy: Chai Discovery's open-source Chai-1 model outperforms Google's AlphaFold 3
“We're lucky to work with the folks at Chai Discovery who just released Chai One, which is open source model that outperforms Alpha Fold Three.”
Guo: Breaking 10% on SWE-bench went from impossible to accessible in 2024
“If you recall, like a year ago, the point of view on sweet bench was like, it was impossible to surpass. Team percent or so. And I think the whole industry now considers that if not trivial accessible.”
Reddy: Test-time scaling currently works best for verifiable domains like math
“It seems like OpenAI has cracked a version of this that works, and we think A, Foundation Model Labs will come up with better ways of doing this, and B, so far it largely works for very verifiable domains, things that look like math and physics and maybe secon…”
Reddy: Mega model labs absorbed $30B-$40B of 2024 AI venture funding
“If you break out the numbers here a bit more, the red is actually just a small number of foundation model labs, like what you would think of as the largest labs raising money, which is upwards of 30 to forty billion dollars this year, and so the reality of the…”
Guo: Test-time scaling allows AI startups to pass compute costs to customers
“I think an underappreciated impact of test time scaling is you're going to better match user value with your spend on compute. And so if you are a new company that can figure out how to make these models useful to somebody, the customer can pay for the compute…”
Guo: Cheaper AI code generation will increase software volume, not replace developers
“If we take the cost of software and high quality software down two orders of magnitude, we're just gonna end up with more software in the world. We're not gonna end up with fewer people doing development.”
Guo: Seat-based SaaS incumbents face an innovator's dilemma from AI automation
“There's a really strong innovator's dilemma. If you look at the SaaS companies that are dominant, they sell by seat. And if I'm doing the work for you, I don't necessarily want to sell you seats. I might actually decrease the number of seats.”
Guo: Incumbents lack the reasoning trace data needed for AI automation
“And then one disappointing learning that we found in our own portfolio is no one has the data we want. In many cases, right? So imagine you are trying to automate a specific type of knowledge work and what you want is the reasoning trace all of the inputs and …”