Welcome back to Forests Over Trees, your weekly tech strategy newsletter. It’s time to zoom-out, connect dots, and (try to) predict the future.
Here’s the plan:
Tech News Takes — super-short analysis and commentary
Strategy Tips — strategy nuggets (for business and life)
F/T Shoutouts — sharing launches, tech events, and other reads
⚡ Tech News Takes ⚡
ChatGPT diminishing marginal upgrades
What’s up: OpenAI’s model improvements are flattening. Apparently, employees told The Information that their next model, codenamed Orion, seems to be less of an advancement than GPT-3 to GPT-4. In response, they are 1/ researching how synthetic data can replace shrinking training data, and 2/ testing ways to boost model performance post-training.
So what: If true, this is a big deal. The astronomical AI company valuations and future-of-work debates hinge on growth continuing to ramp-up. Not to mention OpenAI’s future competitive prospects. Also, as Ben Thompson has pointed out, all OpenAI’s recent leadership turnover has been in the techy, research-centric category. Perhaps they’ll regret being so consumer focused, if it meant neglecting the R&D required to stay ahead.
TSMC breaks rules with China
What’s up: The US ordered TSMC to stop sending its most advanced chips (7nm or smaller) to China. This came on the heels of the discovery that China’s Huawei had somehow acquired such chips, skirting US sanctions. TSMC has agreed and is also now auditing its entire supply chain to avoid similar mishaps.
So what: I was surprised to see TSMC so apologetic, because I thought they wanted to maintain a healthy balance between two enormous powers (China and the US). But in the report by the NYT, they quoted that TSMC’s China-driven revenue was down from 20% a few years ago to 12% last year. So it makes sense to see Taiwan jumping to quickly make amends with the US, especially with a “US first” Trump administration set to return to the White House.
Yelp Buys RepairPal
What’s up: Yelp is buying RepairPal for $80M. The site helps customers get car repair estimates (and helps car repair shops get leads), and the deal is expected to close by end of year.
So what: This is a good move for everyone involved. RepairPal gets a great exit, in an environment where M&A is virtually nonexistent (cough, FTC scrutiny) and everyone is doing acqui-hires and partnerships instead. And Yelp gets to learn further into their strategy of focusing on services (where revenue is growing at 11%, vs overall revenue growth of 4%).
AI beating benchmarks
What’s up: Most frontier AI models are now getting 90%+ on all existing tests of intelligence, making it difficult to differentiate performance. Meanwhile, without harder standardized tests to use, AI companies are developing their own, making comparisons between companies even harder. So to combat that, researchers are developing novel tests to benchmark the latest models, then immediately discarding the tests, to avoid skewing results.
So what: While it might be true that model improvements are slowing (per our earlier story), it’s fascinating to see our testing and benchmarking industries being forced to improve too. Remember the debates about grade inflation, “teaching to the test”, etc.? AI is accelerating problem discovery and solving here. Pretty cool. It also begs the question, how long until we humans can’t even understand the questions AI is being tested on, let alone answer them?
🛠️ Strategy Tips🛠️
Yelp Going Vertical
This week’s strategy session is all about going vertical.
Specifically, I want to unpack Yelp’s acquisition of RepairPal, looking at it through the lens of vertical integration.
But first, Yelp 101.
Yelp has been around for 20 years. That’s 140 in dog years, and a millennium in tech. From the beginning, they’ve focused on discovery, helping users find great businesses, restaurants, etc. My wife swears by Yelp reviews.
But recently, they’ve started to focus less on pure discovery and more on action. Rather than just showing you a map with ratings – and saying bon voyage while you navigate the choppy waters of decision-making from there – they want to stay with you on that journey.
That’s where today’s framework, integration, comes in. There are two flavors.
Horizontal integration is when you buy-up competitors in your same industry. For Yelp’s traditional discovery business, this might mean buying Google Maps (holy behemoth, Batman!... not gonna happen).
On the other hand, vertical integration means buying up your suppliers, or subbing-in for your immediate customer to more directly serve end users.
For my visual learners – ze visual!
In Yelp’s case, buying RepairPal is vertical integration – a move down the visual. This puts them closer to the end users.
RepairPal gives its users a car repair quote that local shops (RepairPal-certified partners) will honor. This saves users from needing to call and heckle several shops, and raises their confidence in the available options.
So if the previous customer journey for a Yelp customer was…
Fender bender
(on Yelp) Research shops, request quotes from 10, hear back from 3, lose confidence as the prices vary wildly, panic, ask your spouse if it really needs fixing…
Pick a shop
Go to shop
Pay
…In the new world, Yelp can solve the pain in step 2, get added to step 3, and get in on the action in step 5! This means they’re involved in transactions (taking a cut), not just running ads.
But the cherry on top is that this move from discovery to action helps their discovery business too. The bookings data and the reams of reviews from RepairPal can make Yelp’s discovery products even better for customers.
Nice job here going vertical.
🌲 F/T Shoutouts 🌲
SAS Loves Synthetics – We covered OpenAI’s synthetic data play earlier, but they aren’t the only ones testing the waters. SAS made an acquisition, and this press release about it has some good nuggets about what synthetic data is and why it matters.
A burden indeed – This is fascinating. Also, ouch (I’m in CA).
