If you're going to call all of that not substantial improvement, we'll have to agree to disagree. Certainly it's the most rapid rate of improvement of any tech I've personally seen since I started programming in the early '00s.
5,217 karma · joined April 18, 2008
Engineering @ Homejoy (YC S10)
hnchat.com:8pJMixG6jypNc7WE7714
If you're going to call all of that not substantial improvement, we'll have to agree to disagree. Certainly it's the most rapid rate of improvement of any tech I've personally seen since I started programming in the early '00s.
While these were almost always very easy tickets to do, they were just one more interruption for us and a blocker for the stakeholders, who liked to have an extremely rapid iteration cycle themselves.
GTM was a way to make this self-service, instead of the eng team having to keep this updated, and also it was clear to everyone what all the different trackers were.
They also report disengagements in California periodically; here's data: https://www.dmv.ca.gov/portal/vehicle-industry-services/auto...
and an article about it: https://thelastdriverlicenseholder.com/2025/02/03/2024-disen...
1. Write a novel that wins the Pulitzer Prize.
2. Prove (or disprove) the Riemann Hypothesis.
3. Provide a theory unifying quantum mechanics and gravity.
4. Design an experiment to give evidence for your theory in (3). The experiment should be practical to actually execute, using no more than the budget to create the LHC (~$4.5 billion).
5. Given programmatic access to a brokerage account with all the permissions of a typical hedge fund, raise all the money required for your experiment in (4) by trading on the stock market, starting with $100.
6. Solve for (5), without being provided access to an account first - begin with just a general internet connection and use computer security vulnerabilities (known or zero-days that you discover) to get some way of trading instead.
7. Solely by communicating over the internet, establish a new religion, and convince at least 10 million humans to convert to it. Converting should require adherence to a strict code of conduct that a random, unbiased panel of human judges consider to be at least as strict and challenging to follow as the tenets of Hasidic judaism.
8. Implement an AI which could score higher than you on questions 1-7 with lower total cost of compute.
There is also a much bigger group of people that haven't really tried anything beyond GPT-3.5, which was the best you could get without paying a monthly subscription for a long time. One of the biggest reasons for r1 hype, besides the geopolitical angle, was people could actually try a reasoning model for free for the first time.
a) For inference, cheaper and faster compute will increase total inference spend, because the end-user products will work better and people will use them more.
b) For training, the big labs will continue to spend because we have yet to see diminishing returns to scale - in fact, we have in the past year unlocked a new dimension to scale up training-time compute - doing more RL after pre-training to improve reasoning capabilities. Since current SOTA models are not yet smart enough for all the tasks people want to use them for, this means that any efficiency gains will be used to further improve performance. In the current competitive environment, even with DeepSeek's work, it's near-impossible to imagine OpenAI, Anthropic, Google, or Meta deciding to cut the compute budget for training their next model by an order of magnitude. They will still incorporate DeepSeek's techniques into their next model, but use them to squeeze even more performance out of the compute they have, and will keep purchasing as much compute as NVidia will sell them. Expect this trend to continue until there are no more returns to scale anymore.
The cleanest way for this to work is the for-profit to just sell more shares at the $150B valuation, diluting the non-profit entity below majority ownership. The for-profit board, which the non-profit could still probably have multiple seats on, would control the real asset, the non-profit would still exist and hold many tens of billions of value. It could further sell its shares in the non-profit and use the proceeds in a way consistent with its mission.
They wouldn't even have to sell that much - I am pretty sure the mega-fundrasing rounds from Microsoft etc brought the non-profit's ownership to just north of 50% anyway.
I don't see how this wouldn't be above board, it's how I assumed it was going to work. It would indeed mean that the entity that controls ChatGPT would now be answerable to shareholders, a majority of which would be profit seeking and a minority of which would be the non-profit with its mission, but non-profits are allowed to invest in for-profits and then sell those shares; all the calls for prosecutions etc seems just like an internet pitchfork mob to me.
I have found the opposite to be true, starting this year. You can commiserate with people who are interested in cryptography.
IMO this hasn't changed as much as it seems from the outside. The brilliant individual leaders of the past owed most of their success to their team members, and most high-functioning teams today, if you study how they work internally, do actually have a key leader who is aligning everyone around a shared vision. What's changed is that it's no longer seen as a positive to boost the profile of the leader too far the rest of the team.
https://www.bls.gov/opub/ted/2023/union-membership-rate-fell....
What we're talking about is erecting regulatory barriers for a technology that could solve one very large problem (auto safety), just because that technology doesn't solve some other problem (climate). This is like saying data privacy protections on a social networking site are not the right solution, because they don't really solve the misinformation problem.
My parents who live out in the burbs can only drive for so much longer. I sure hope they aren't as stubborn as my grandparent who drove to the age of 90 and got into an accident (fortunately with only a parked car), but either way I don't think finally learning to ride a bike at the age of 70 is the answer either.
We should move forward with rolling self-driving cars as quickly as we safely can. Speaking as someone who does not own a car and has no desire to ever own a car, I think it would benefit me a lot too, by making it less likely for me to get run over while I walk or bike around, the fear of the latter being the biggest obstacle preventing most people I know from biking more too.
Of course, the difference is that the technologies causing climate change are more deeply embedded throughout the economy, and so political resistance to anti-climate change measures is very strong as well. This is one reason in favor of addressing risk earlier, before we make our civilization as dependent on large neural nets as it currently is on fossil fuels. A climate change movement in the mid-1800s when the internal combustion engine was just taking off would also have been seen as quixotic and engaging in sci-fi fantasies though.
In the scenario where the current wave fizzles out and we have another AI winter, one risk is that we'll be left with a big regulatory apparatus that makes the next wave of innovations, the one that might actually get us all the way to an algined-AGI utopia, near-impossible. And the regulatory apparatus will now be shaped by an org with ties to the current AI wave (imagine the Department of AI Safety was currently staffed by people trained/invested in Expert Systems or some old-school paradigm).
I can say that Google and Microsoft would definitely care, just as two specific instances of "anybody" that I have worked for and seen what their legal review process gating new product launches is like.
You seem much more optimistic that the community efforts are <6 months behind those companies than I am; I would be happy to be wrong here and see an open-source self-hosted LLM with better than GPT4 performance by September.
It's a very exciting, multi-generational project to think about. Though frankly I think we're going to get AGI + brain/computer interfaces + cloud mind uploads before we get martian terraforming; and dropping the bag-of-meat related requirements would make extraterrestrial colonization much easier.
The manual "reporting/fixing" of issues, which was done over the phone, includes pilots and crew members reporting "hey, I didn't actually get on the flight I was scheduled to". If the system did not receive that message, it was built to assume that the crew members made it to their next location and would be available for their next scheduled flight. But once a certain threshold of flights got canceled because of the storm, the limiting factor became how quickly staff members could even report that they didn't make it to their next location - they'd stay on hold for hours, until their next flight in their next city was scheduled to leave, and the system would still assume they were on that flight because it hadn't yet heard otherwise.
Lack of webapps certainly wasn't the only issue here - as the article notes, their whole point-to-point scheduling model is more vulnerable to cascading failure as well, vs. a hub-and-spoke model. But the phone-based system did certainly cause a feedback loop that kept the disruption happening for much longer than it should have. And moving away from point-to-point would have many more tradeoffs/downsides for consumers than simply modernizing their tech system.
1 - The companies that make the game. Whether it's Riot and Blizzard selling slots in their leagues for eight figures, Valve using their annual tournament to sell in-game cosmetics, or all the companies ultimately owning broadcast rights to their game, this is the biggest difference between esports and traditional sports.
The New York Yankees and Los Angeles Dodgers make money because they sell their own tickets and broadcast rights to their games; even though MLB does control the streaming revenue, they share it out to teams and ultimately have to bow to the wishes of a majority of team owners. What you don't see is a single company MLB inc that owns the copyright to the game of baseball, sells broadcast rights to all teams themselves, charges teams to play in the league, and can kick teams out of the league at a whim. That's the situation in esports.
2 - Individual players, with streaming. Players can first make a name for themselves in competitions, then stream on their own Twitch channel for revenue. This is not that different from athletes acting as social media influencers and signing endorsement deals, but the biggest difference is that by streaming on Twitch, they appear side-by-side with tournament broadcasts. It's as if LeBron's instagram account where he streamed his workouts and pickup games were just one change-of-the-channel away from ESPN, and people would consider it normal to flip between the game and individual player streams.
Lots of esports orgs, as part of signing players, get a big cut of the player's streaming revenue. But the revenue for an individual player's twitch stream, while great for an individual, usually isn't going to be significant enough to maintain a whole organization, and when a player brand does get big enough that their stream could sustain an org - that's when the player will be heavily incentivized to go independent, and make more money from streams than they do from competing.
Ultimately, I think esports has a bright future - overall total viewership continues to rise, even though some games like League of Legends - which is more than a decade old now - are starting to fade. It's just the business models of the offline sports world don't carry over, and that's especially apparent with these organized teams.
The alternative world, where we're using crypto to pay for everything, and we're no longer logging into twitter, but instead using our keys to log into a client for a globally distributed twitter where we own all our data on a blockchain - that's impossible without not just Eth2 but a several orders of magnitude of improvements beyond that.
Basically, I think taking the current per-transaction carbon impacts, multiplying it by eg. Visa's transaction volume, and forecasting a world where crypto is burning 1,000X as much carbon as the whole transportation sector or whatever - is a fallacy. Crypto will either solve the scalability problem (and therefore consume a much more moderate amount of energy), or it won't be used at all.