liamconnell

I think NYT uses multiple titles for some articles. I had copy pasted it

I’m guessing that Amodei meant it as a humorous inside joke.

It’s also shorthand for “the end of massive R&D capex” and “the transition to market capture”. The final stage, what McKinsey types call “harvesting”, is probably not on Amodei’s radar. Based on what I’ve seen of his public personality, he would see it as too philistine and will hand it off to another custodial exec.

It is, but when a model/harness/tools/system prompts are the same/similar in the generator and reviewer fail in similar ways.

Is there empirical evidence for that? Where is it on an epistemic meter between (1) “it sounds good when I say it”, and (10) “someone ran evaluation and got significant support.”

“Vibes” (2/3 on scale) are ok, just honestly curious.

Not mentioned in the article, but one interesting area where OpenAI could play is in participant identification and recruitment for clinical trials. In fact, ChatGPT could also help operate the clinical trials which is a highly paperwork intensive business, and therefore something that AI could add value to.

Ultimately pharmaceutical companies pay up to $100,000 per participant to hospital networks these charges must be itemized as expenses from the hospital on the most part (bounties are illegal usually.) open AI would provide a cheap way in for pharmaceutical companies to identify participants given that OpenAI has an incredible perspective into the physical and psychological state of their users. Imagine how much more is shared with OpenAI compared to a clinical trial coordinator at a hospital when a psychiatric drug is being tested.

This would also give OpenAI leverage in partnering with pharmaceutical companies. OpenAI executives have stated this is a goal, but otherwise they’ve made little progress on it.

It’s wild to imagine - someone with borderline personality disorder having delusional conversations with an AI chat Bot for six months, receiving an offer to participate in a clinical trial, and then having their subsequent AI conversations used as evidence to analyze the efficacy of the drug. The ironic thing is if that person had delusions about hidden forces listening to them…they’d be RIGHT!

Someone would purchase it and run it again under a new company.

That happened a long time ago! Microsoft already owns the model weights!

Loved Bluebeard as well. A mature Vonnegut who knew how to use motifs from his earlier work. And for an old guy, he kept his writing fresh and energetic. The miniature story of the dog without a tail always comes back to me.

I read the second two books this year and was unfortunately disappointed.

First two were so fun but I think I got hung up on some of the more clumsy stylistic parts of the Endymion books. I guess my “trust” in the author comes from the style and tropes they use, and if I they lose my trust none of the deeper parts resonate. Glad you enjoyed!

Neither are on-brand for hacker news but:

[Best fiction] Stoner - John Williams [Best non-fiction] Say Nothing - Patrick Radden Keefe

The US crypto lobby has already raised over 240 million for the 2026 election. It’s a cancer attacking our society’s institutions.

Didn’t see that yet, interesting!

The CFO isn't talking about selling tokens to pharma companies. There's no money in that. She proposed revenue sharing. In this scenario, OpenAI's AI service helps discover drug candidates, and shares the IP ownership of the candidate (which is basically a risky bet that it will get through clinical trials and be profitable). Biotech is a complicated market filled with smart people and great negotiators - they dont give away IP ownership without a lot of thought.

If OpenAI wants anything more valuable than selling tokens, they will need to offer something valuable and differentiated. Right now they are not differentiated in the space at all. Look up "OpenAI Biotech" - anything that they've built themselves?

If any company will have a new product that biotech companies will pay top dollar for, its Google. Deepmind has been in biology (proteins) for almost a decade and they it has subsidiaries like Isomorphic Labs that are bringing products to market.

Good point!

On the other hand, Historically Amazon didnt compete with Google (until GCP). They do compete with Microsoft, which is pretty closely aligned with OAI. They also have large investments in Anthropic.

Even if OpenAI did win here, would it be a profit monster like Google Adwords? Adwords had the auction model which meant that certain categories were hugely lucrative for Google. Can a chatbot do the same? If I know that the product I buy is simply auctioned off to the highest bidder, what's the point of using an agent to help me shop? There has to be a pretext of the agent actually looking out for my best interest, otherwise I would just use search. Nobody expects adwords to look out for their best interest. They are always free to skip the ads section if they choose.

It will be hard for ChatGPT to implement an auction model since it will be different for each product category. Hiring a lawyer will probably have a different interaction from buying groceries. On Google+AdWords, its all just search results and ads.

If there is no auction, then all of this is WAY less profitable than the Google model. So once again - not going to save OAI from negative margins.

Yes that's what I meant. But Amazon will fight for its life to stop this. So OpenAI will have to go to other retailers, which don't have the same product catalogue size as Amazon. Best case scenario: Target or Walmart. But there is a reason that OpenAI announced deals with Etsy and Shopify rather than those.

And OpenAI doesn't have as much product insight as the retailers so they have to rely on the retailer to choose which is the "best" mechanical keyboard for this person. And at that point, pretty much all of the shopping value is being provided by the retailer rather than ChatGPT, so why would they get much money?

There's a market for this but its not going to be trivial for OpenAI to win it. And it probably wont be a cashflow monster like AdWords or Amazon.

Yes but its not relevant to the agent use cases (which are mostly about interacting with external systems). So agents built by Microsoft (Copilot) can natively interact with Office files in Sharepoint, and the Sharepoint product team can build to enable this in special ways. OpenAI has to use the APIs and deal with rate limits, speed issues and other limitations.

Unfortunately for OpenAI, they are not positioned to capture value from any of the "big margin" use cases that they highlight as key to their future. I think all of these are pretty unrealistic for them:

- Revenue sharing from drug discovery (called out by OpenAI CFO): Why would a pharma company give away the upside to a commoditized intelligence layer? Why would OpenAI have a more compelling story than Google Deep Mind, which has serious accolades in this space?

- Media generation for ads and other content: For ads, OpenAI is facing off against Google, Meta and Amazon, all of which have existing relationships with advertisers. For the foreseeable future, AI content will be a major discount product compared to humans. OpenAI will not get to charge $1M for an ad like a production company does. So the TAM of ad production (~$50B) shrinks below $1B because AI deflates prices so much.

- Other agent use cases: OpenAI doesnt have a surface to build these on. Google has chrome, Microsoft has office, Apple has OS's. The other use cases like coding will be a low-margin competition between model providers until some of them throw in the towel. The players with the best cash position win - and thats not OAI.

I think the place that they could win is retail (also called out by OAI CFO). They made deals with Etsy and other small retailers. I was fixing my guitar the other day and would have instantly bought the tools it had suggested that I would need. The problem is that they have to win against Amazon here, and there is zero chance of a partnership for obvious reasons.

I usually play piano improvisationally, and manually hitting record never meshed well with that.

So relatable.

Congrats, looks like a great product. I just ordered one for my piano-player buddy for Christmas.

Jane street did it for a long time. They are quite large now and only recently started bringing in program managers and the like.

Well I’m not saying I agree with it, but a comparison to the Great Depression is not out of context for an article comparing today’s economic situation to that immediately preceding the Great Depression!

Do the private companies get some special IP rights on the public sector research? It seems like in a competitive market, those private companies would have thin margins. What stops a lower cost competitor from using the same public IP? I’m clearly missing something important here.

I lived in Argentina for a few years. There are plenty of problems with their public healthcare system that pretty much mirror any country with a troubled economy, but one day a work colleague recommended I visit the Favoloro Institute. I’ve never seen such an impressive healthcare organization in my life.

They time their appointments ever 2 minutes so you’re expected to arrive early. Once you start, they wiz you around to about a dozen tests including getting hooked up to electrodes while running on a treadmill. After less than an hour you see a doctor who goes through everything for about 15 very detailed and thoughtful minutes.

The entire thing is FREE! Completely publicly funded.

I don’t know much about healthcare but this type of thing always struck me as a missing apparatus of healthcare in the US. Although apparently Favoloro (who invented some kind of bypass surgery) founded the Institute after being inspired by the Cleveland Clinic.

Would love to hear other people’s thoughts on this because I tell that story all the time!

Gpt2 answers to gpt3. Gpt3 answers to gpt4. Gpt4 answers to God.

If you study history you’ll notice that groups and their leaders are rising and falling, conquering and pillaging and then losing it all. The world is too dynamic for your reductive theory to fit in.

Elites compete with each other. They don’t sing kumbaya and cooperatively share the keys to power.

Well hold on now. Parent presented that as fact with no evidence too!

I think you can integrate it with nginx, which would work well with postgrest.

Not sure why you were downvoted. Parts of the book are indeed timeless!

My favorite but is the idea that out of the jumble of documents produced, a few emerge as the pivot around which the majority of the team’s communication revolves.

I’ve seen very effective project leads use this to great effect. Rather than robotically using boilerplate document templates they discover the most precise format for the project at hand.

Great point. I have a comment below that goes into some of the difficulty with this distinction.

To be fair, many small/mid-sized companies that act as suppliers to larger companies would basically have the larger company doing the carbon accounting for them.

For large companies, I think it would look more like an accounting capability than a climate/energy science one. The carbon credit market might become regulated with some accreditation system after these wild west years.

government sequestration sites

Well that would drive up demand for fossil fuels and profits for extraction companies, leading to more extraction. Is this a real idea (proposed policy in a country) or just something you wrote down?

Great point. This would be great for a net importer country like the US. Germany (net exporter) wants their goods competitive abroad.

On the other hand, if the US and other high-consumption countries collaborated on this it would probably be effective.

Like I said, I go back and forth on it. Or better put, I think both approaches are valid albeit with big implementation challenges.

I go back and forth on the push for carbon neutrality for large companies (as opposed to some form of carbon tax). I'll give a few examples that I've seen first-hand in my work as a consultant.

1. A large state-owned African agricultural company is using satellite and soil-sample data to measure the carbon sequestration potential of agricultural land if certain farming techniques are used (no-till, etc). They use satellite and soil-sample data feeding into a biological model implemented in fortran (from some academic paper that may or may not be replicated) for the estimation. The plan is to go to farmers and ask them to switch the technique. The company would sell carbon credits, give the participating farmers a cut and keep the rest. This has great potential! but the large-scale implementation is very difficult manage. The company would at least need to randomly check in on participating farmers. It should also take soil samples over time to make sure the technique is working. Carbon credit sellers should be auditable, but the government relations and size of the company might make that difficult to do rigorously.

2. An analytics tool hopes to help large companies identify the parts of their supply chain that can reduce the overall carbon emissions of a product with the lowest price increase to the end-consumer. Think about BMW sourcing steel from China for example. If this helps companies actively manage their supply chain in order to lower emissions, then suppliers will adopt to low-carbon techniques over time. The problem is that it requires suppliers to disclose their emissions accurately. Many supply chains in carbon-intensive industries are global and fragmented. If a bad-actor supplier on the other side of the world lied and got caught, they could just change their name and carry on.

3. Global supply chains also present issues with carbon taxes at the drilling/mining site. You can't force foreign countries to tax this (and if you did they would be incentivized to cheat), and you can't force foreign companies to accurately disclose energy use. A domestic energy tax would make local manufacturing uncompetitive, which is something most countries are not willing to sacrifice.

The debate between carbon neutrality vs carbon/extraction taxes is the choice between inventing an entire new bureaucracy of enforcement (carbon audits, carbon disclosure, etc) or the impossible task of rallying and enforcing global action (every country imposes the same tax, no exceptions). As an engineer, I understand the seeming beauty and efficiency of the second option, but I think the first is ultimately going to be our best bet.

Krugman also claimed in the same article that Capital in the 20th century was not fully read by most "Like A Brief History of Time".

He is right that Piketty's first book is probably just worth it for the introduction and most non-economists never read the rest, but if he cant get through A Brief History of Time then he's got other problems. Hawking made that book very readable and accessible!