HN user
i do money stuff
Young man me likely would have thought, “wow, cool tradition!”
Old man me thinks “$390 million? How are they funding this?! That seems like a massive sum of money to throw down the drain every 20 years.”
Then I did the back of envelope math. Assuming 20% comes from donations, then all you’d need is a $380m fund earning a real 3% to fund the building of the next temple. That’s very doable.
My main takeaway from this is that Luca Maestri was a blight on Apple. He steered this company into the most malicious of compliance schemes that only has increased their regulatory attack surface. On top of that, he's the one who discouraged apple from making huge investments in compute for LLMs that threatens to derail Apple and the iPhone's primacy as the gateway to information.
You’re only looking at one side of it. Consider how much value that excel adds, and how many use cases it enables. It’s incalculable. It’s always easy to poke at excel’s issues, but errors are almost always a skill issue.
From a financial perspective, there are many ways you could enforce checks to ensure the model is balanced - it just takes time. Data entry can be an issue, but you can automate that too. The deloitte report is saying the health department should benefit from adopting a gigantic erp system but you could get 90% of the benefit by employing a couple people that really know what they’re doing.
You could say that excel should allow those things to happen but the flexibility is precisely what makes it so valuable.
The people writing the reports are consultants. Consultants recommend things that benefit consultants. In their case, a multi year process and tens of millions of dollars trying to install ERP software is more a windfall for the vendors and not the companies.
Apple outsourced chips to IBM (PowerPC) for a long time and floundered all the while. They went into the game themselves w/ the PA Semi acquisition and now they have Apple Silicon to show for it.
If Apple wasn't selling privacy, I'd assume the other way around. Or if anything, OpenAI would give the service out for free. There's a reason why ChatGPT became free to the public, GPT-4o moreover. It's obvious that OpenAI needs whatever data it can get its hands on to train GPT-5.
I can't really dispute any of that.
It took almost a decade but the PA Semi acquisition showed that Apple was able to get out of the shadow of its PowerPC era.
Nvidia will remain a leader in this space for a long time. But things are going to play out wonky and Apple, when determined, are actually pretty good at executing on longer-term roadmaps.
Nvidia obviously has an enormous, enormous moat but I do think this is one of the areas in which Apple may actually GAF. The rollout of Apple Intelligence is going to make them the biggest provider of "edge" inference on day one. They're not going to be able to ride on optimism in services growth forever.
Learning how to write llm function calls.
Can't wait to attempt GUI automation on my iPhone using the mac.
If we know anything about Apple, they're going after Nvidia. If anyone can pull it off, it's going to be them.
Apple Intelligence stuff is going to be very big. iOS is clearly the right platform to marry great UX AI with. Latching LLMs onto Siri have allowed the Siri team to quickly atone for its sins.
I think the private compute stuff to be really big. Beyond the obvious use the cloud servers for heavy computing type tasks, I suspect it means we're going to get our own private code interpreter (proper scripting on iOS) and this is probably Apple's path to eventually allowing development on iPad OS.
Not only that, Apple is using its own chips for their servers. I don't think the follow on question is whether it's enough or not. The right question to ask is what are they going to do bring things up to snuff with NVDIA on both the developer end and hardware end?
There's such a huge play here and I don't think people get it yet, all because they think that Apple should be in the frontier model game. I think I now understand the headlines of Nadella being worried about Apple's partnership with OpenAI.
Regarding Germany and large corporations, and somewhat of a tangent, I remember a decade ago a bunch of hedge funds had tried to sue Porsche, the parent company of VW, for cornering the market for VW’s open interest and cause the mother of all short squeezes.
They tried the case in New York but it got thrown out for lack of jurisdiction. They did try the case in Germany, but Porsche had fittingly cornered the market for the best and biggest law firms. All of the best law firms refused to take the case because it would mean that they’d be essentially blacklisted by the largest companies in Germany for bringing a case against a German company.
It’s taken a decade, but I now see a pattern.
There weren’t 200 holdouts. It was like 5 AM over there. I don’t know why you are surprised that people who work at OpenAI would want to work at OpenAI, esp over Microsoft?
Likely connected to the fact that they released voice chat for free users an hour ago. https://x.com/openai/status/1727065166188274145
Assuming these two things are related, if I may editorialize just a tiny bit, I am a little annoyed at how much their rollouts often disrupt service for paying customers. Paid users being impacted by free user rushes really sucks, but is understandable. API developers being impacted by free-user rollouts is unacceptable, and especially sucks for those who have to answer to users of their own.
I suppose this is a wakeup call to migrate to Microsoft's Azure endpoints which, presumably, aren't affected by the current outages. But I'm fully tapped out in terms of yet another service's application and vetting process.
So to connecting it back to the current drama, while I support OpenAI, their employees, and Sam's return, I can understand why folks like Helen would be miffed by management's approach to building. I'm not saying they should slow product development, but would staged rollouts hurt?
Don't forget Tasha McCauley's husband, Joseph Gordon Levitt, has been vocally anti-AI during the SAG-AFTRA strike, an event for which AI was a huge point of contention. It's a poisoned board.
Seems all but confirmed that Sam is going back to OpenAI.
You have! The Apple Vision Pro is slated to be released by the March/April 2024. The device has only been made possible by the steady march of Apple Silicon.
Saying the chips haven't enabled any sort of innovative products it's like driving down the interstate wearing goggles, err, blinders. Let's check back in four months.
I do think he was the chief architect of the coup. I do think his beliefs and ideals are still valuable flora for a company of this ambition. There just needs to be a more professional structure for him to voice them.
Dealing with folks like Ilya isn't necessarily a matter of if, but how much.
I think that his beliefs are important to the company. A board shouldn't be a homogenous glob nor should it be like a middle school friend group. What he did was both bizarre and amateur, but I believe in the best of us all the come forward from these types of events.
I don’t understand why and how they didn’t consider this sort of discussion before so unceremoniously firing him. The others on the board outside Ilya need to go.
I don’t consider anybody beyond forgiveness and if Ilya takes a professional lesson from this and Sam learns to be more mindful of others’ concerns, I consider this a win for all. Starting over in a new entity sounds great but would be years of setback.
I hope they work this out.
ExifTool[0] does all that from the command line. I use it for automating my photo organization workflow and, as a bonus, I use it for reading the metadata of damn near any filetype.
My interpretation is actually the opposite.
I believe that 1Password has grown significantly in the professional space, and that Okta likely sees this trend as a competitive threat.
If there's anything that Apple's upending of Blackberry taught us, it's that all you need is a one key decision maker in an entire organization to say "wait, why can't I use {product x}?" for that organization to slowly move away from the entrenched enterprise solution. The fact that Okta is joining the consumer space validates 1Password's decision to creep into the enterprise market.
I'll push back because, in my experience, this sensation arises in the still-very-tired aftermath of totally spending myself on a problem.
Solutions emerge in my head as I sleep, usually after a day of brute-forcing a problem so much that it spills over into my mind's eye during breaks, which spill over into my dreams, where glowing forms congeal into solutions.
I recently read some comments here about people having vivid dreams of playing Tetris. That's exactly how my dreams feel, but in mine, I see either code, foreign languages, or a spreadsheet. Extending your gym analogy, it's like hitting your PR the day after going all-out on volume training.
No such suspicion is necessary. Google's revenue from the United States has only increased as a share of its total revenue over the past decade. https://abc.xyz/assets/4c/c7/d619b323f5bba689be986d716a61/34...
Keep in mind that the information in the link you provided only applies to the developer models, which are only accessible via API and OpenAI Playground.
OpenAI's developer relations person confirmed it a little whiles back:
[..] I made the comment elsewhere, but the models are always changing in ChatGPT.
https://twitter.com/OfficialLoganK/status/166447707026357043...
GPT-4 in ChatGPT has changed so much and so frequently. I've spent an inordinate time using both, side by side, and it has mostly failed to replicate the performance of their API offerings. While I don't have examples on hand right now, you can also look for yourself. Just look at the network requests in the web inspector. Here's the description of the Default ChatGPT GPT-4 model:
{
"slug": "gpt-4",
"max_tokens": 4095,
"title": "GPT-4",
"description": "Our most capable model, great for tasks that require creativity and advanced reasoning.",
"tags": [
"gpt4"
],
"capabilities": {},
"product_features": {}
}
It's really poor work.
There are two conclusions I took from scanning through this and trying to reproduce a few of the reported failures.
1. The author is bad at prompting. There are many ways to reduce hallucinations and provoke better thinking paths for the model.
2. The author is using ChatGPT's GPT-4, leading him to conflate "GPT-4" with "ChatGPT". While you can consider this a shared failure with OpenAI, due to OpenAI's poor communication, anybody doing serious work evaluating these models would know that the first thing you need to do is use the API and pin the model version. In the author's case, he should have used gpt-4-0314 or gpt-4-0613. What I suspect he did is that he just used ChatGPT's GPT-4, and likely the default model at that. (Nobody should ever use the Default model. It's their most heavily performance optimized model and performs worse on reasoning tasks than the Plugins model, even on within-context-size tasks.)
There are huge problems with that, because OpenAI has done both a ton of fine tuning and performance optimization continuously on the default ChatGPT model over time that its performance has ranged anywhere from "I'm pretty sure this is gpt-3.5" to "whoa, this is damn good" (the latter being mostly the model at launch, which was probably the same as gpt-4-0314).
If the author has been working seriously at evaluating models, specifying the model is the first thing he'd do. Perhaps he should explain his reasoning.
Median income is not a better measure because it is blind to changes in population composition over time. Series of median income figures doesn't actually track anything.
For instance, a country that accepts more lower-skilled immigrants will see its median income figures grow less than countries that accept fewer of those immigrants. Is the immigrant-accepting country poorer as a result?
The only way that median income would be a valid measure of wealth or QoL would be if you isolated the cohort of those considered "median" in 2008 and tracked their wealth over time.
Since last year, before I heard about langchain, I've been building my own stack of tooling for my own LLM projects that probably now covers about 10-20% of Langchain's functionality. I heard about Langchain earlier this year and groaned, thinking that I did a lot of work for nothing..
..Then I actually used langchain. I was shocked at how poorly performant the code is. Some operations took 10x longer than how I did it, and all the while producing worse results. As tempting as it is to just roll with langchain from day one, I'd highly advise against it. Think deeply about what you're actually trying to accomplish and instead of just injecting langchain in the middle of everything as this messy, amorphous glue code thing.