i helped Chris Callison-Burch design a class at upenn, called interactive fiction, which is a similar context to what Simon suggested. the real magic is that it reframes hallucinations as creative story telling. the usecase is SUPER fun if you imagine the LLM as a dungeon master telling a story that gets expanded over time.
the framework he and I built kept track of the game state over time and allowed saving and loading games as json. we could then send the full json to an LLM as part of the prompts to get it to react. the most neat part, imo, was when we realized we could have the LLM generate text for parts of the story, then analyze what it said to detect any items, locations, or characters not jn the game state, and then have it create json representations of the hallucinated objects that could be inserted into the game states. that sealed the deal for using hallucinations as creative story telling inside the context of a game.
i wasnt officially part of upenn at the time, so my name isnt listed on the site, but we wrote a paper about some of the things we did, such as this one, and you'll see me listed there https://www.cis.upenn.edu/~ccb/publications/dagger.pdf
Sounds similar to AI Dungeon which I believe ran on a fine-tuned version of GPT-2 "all the way" back in 2019. And honestly kind of reminded me of the "Mind Game" in the novel, Ender's Game.
The article is about image generators. Image generators specifically work by starting with noise and then refining the noise into an image. That's not how driving software works and this is not a relevant point.
Sorry, I failed to follow your reasoning. My comment had nothing to do with "driving software", it addressed the parent post by posing the question a different way.
Jobs hated ads. You're right that he never wouldve done what Apple is doing now.
Cook needs to stop listening to investors, like Warren Buffett, because he's letting them wreck Apple's integrity for the sake of making a buck. Apple just isnt user focused like they used to be and it's crappy.
he was vocal about his opposition to intrusive ads in particular. he'd say "You’re either the customer or you’re the product." he believed users paid a premium for apple products and that they should not be subjected to compromises with advertising.
iAd was something that happened right at the end of his life because devs were putting ads in apple apps anyway and he wanted to control how that was done.
this is meant to add context to what bluedevilzn said, btw. it is not a refutation.
Jobs disliked anything where Apple wasn't getting a cut. Flash games and Google ads being two of the biggest offenders in his eyes.
He also "hated" the small tablets Samsung were making, saying in a keynote that you'd have to file your finger down to use it. He said this knowing full well Apple were launching the iPad Mini in 12 months' time.
I really hope one day Jobs' marketer-speak soundbites stop being repeated like like biblical pronouncements. The App Store, Apple News, Stocks and other properties are filled with hideous Google-like ads today, and Jobs likely wouldn't bat an eye, because they brought in money.
I think Jobs recognised that ads are intrusions into people’s lives. The advertiser has a responsibility to respect the audience. They don’t have a natural right to that attention, and have to earn it.
Thats why the F1 wallet add is such a bad move. It’s disrespectful and intrusive.
iAD was supposed to be about innovative, informative, well designed high quality adverts. It never really worked out though.
Yeah, “Jobs hated ads” is a such a wild rewriting of the history of one of industry’s greatest marketers and, yes, ad men. (1984 commercial. Mac vs PC.)
I am curious what you attribute that Warren Buffett is asking Tim Cook to do? Warren is notorious for being hands-off with operations. I can't imagine him having ANY commentary on what Tim Cook should be doing with Apple other than with capital allocation.
Cook is an operations person. He makes the logistics work. He's no visionary. Jobs is a visionary, but is not a logistics person. Apple struck lightning when both existed, to provide complimentary ideas and counterbalances.
Same with Ive and Jobs. Ive was a great designer, but no usability expert. Jobs put practical limits on and as soon as Jobs was gone, Ive got total control. The result is some of the least-popular Mac laptops ever.
depending on what you're using the synthetic data for, it is sometimes called distillation. here is a robust example from some upenn students: https://datadreamer.dev/
Prior to LLMs, it was amusing to consider how ML folks and software folks would talk passed each other. It was amusing because both sides were great at what they do, neither side understood the other side, and they had to work together anyway.
After LLMs, we now have lots of ML folks talking about the future of software, so ething previously established to be so outside their expertise that communication with software engineers was an amusing challenge.
So I must ask, are ML folks actually qualified to know the future of software engineering? Shouldnt we be listening to software engineers instead?
> So I must ask, are ML folks actually qualified to know the future of software engineering?
Probably not CRUD apps typical to back office or website software, but don't forget that ML folks come from the stock of people that built Apollo, Mars Landers, etc. Scientific computing shares some significant overlap with SWE, and ML is a subset of that.
IMHO, the average SWE and ML person are different types when it comes to how they cargocult develop, but the top 10% show significant understanding and re speed across domains.
This seems to be overstating the separation. For people doing applied ML, there's often been a dual responsibility that included a significant amount of software engineering. I wouldn't necessarily listen to such declarations from an ML researcher whose primary output is papers, but from ML engineers who have built and shipped products/services/libraries I think it's much more reasonable.
Is this substantive engagement with the earlier comment? I’m not seeing it. You probably know the examples are different (long term R&D on a telecom protocol followed by government implementation and standards and industry adoption … versus fairly early-days access to a GenAI model tuned for defense contexts).
Generally a notable counter-example to a broad-brush point stands as "substantive engagement", yeah. Stating that the Pentagon buys software badly in the general case is a less specific and less engaged point than "ARPANET and IPv4 were DoD projects that ate the world".
If you want to argue that the examples are different, that's an extra point you need to bring to the table. You're not allowed to assume everyone just agrees with you.
Don’t you think the standard you mention is too low? I do.
The comment didn’t advance the conversation. It was a relatively shallow level of engagement; something I’d expect to see in a silly Reddit back and forth. We deserve better here.
And to your point: my comment explained my point: “long term R&D on a telecom protocol followed by government implementation and standards and industry adoption versus…”.
Of course I don’t assume everyone agrees with me. (You don’t really think I do, do you?) But I want people to put a certain level effort to reach a quality bar. My problem perhaps is that people don’t want to put in sufficient effort. Or perhaps as a community we are not setting the bar high enough. This level of thinking is attainable here; we just need to set the bar and fight for it.
Dunno. I think you're being pedantic about a point that is clearly incorrect. In fact the history of Pentagon-funded R&D is absolutely filled with wild success stories and with embarrassing disasters, as you'd clearly expect from any organization that size.
I don't think you're prior is correct at all here, and trying to dismiss a bleedingly obvious counterexample (I mean, come on!) as "shallow" just because it refutes your deeply held beliefs is exactly they opposite of "substantive engagement".
> Dunno. I think you're being pedantic about a point that is clearly incorrect.
If I was incorrect about a positive (factual) claim, that's fine, I'm happy to learn. However, if we disagree on normative claims (values), that is not about "correctness".
For context, I can't readily think of a time when someone here on HN used the word "pedantic" in a kind way. It seems like the most-socially acceptable form of insult here. It isn't something you say out of respect. Your tone seems angry and combative rather than trying to understand. From my point of view, this is sad and counterproductive. To hoist up a level, this is part of my main point above -- I want there to be a kind of discussion here that is productive: both substantively and in terms of mutual respect.
> the history of Pentagon-funded R&D is absolutely filled with wild success stories and with embarrassing disasters...
I don't disagree. I'm not sure why you think I would. Perhaps you were misunderstanding?
Just to give some context -- which you probably know -- but it will help give us some shared grounding ... DARPA funded TCP/IP by way of the ARPANET, and DARPA gets its funding from the Pentagon. Still, the Pentagon's R&D funding (around $140B) is hugely different in character and scope from DARPA's funding ($4B). Compare (i) a broad Pentagon contract to get access to OpenAI's services with (ii) DARPA's funding for ARPANET. I don't see this as an interesting or relevant comparison: the Pentagon isn't driving fundamental research in (i). There are much better comparison points, such as the Pentagon's contracts with AWS or Azure.
There is no need to be a jerk about it. I explained my standard and thinking; we don't have to agree, to restate your earlier point. Your comment is choosing a "win versus lose" mentality.
That was 51 years ago at about a trillion spent a year since. Have any examples from the 21st century? Keep in mind they also essentially lost every war they fought during that time as well.
the framework he and I built kept track of the game state over time and allowed saving and loading games as json. we could then send the full json to an LLM as part of the prompts to get it to react. the most neat part, imo, was when we realized we could have the LLM generate text for parts of the story, then analyze what it said to detect any items, locations, or characters not jn the game state, and then have it create json representations of the hallucinated objects that could be inserted into the game states. that sealed the deal for using hallucinations as creative story telling inside the context of a game.
i assure you the D&D context is very fun! the class website might give you more ideas too https://interactive-fiction-class.org/
i wasnt officially part of upenn at the time, so my name isnt listed on the site, but we wrote a paper about some of the things we did, such as this one, and you'll see me listed there https://www.cis.upenn.edu/~ccb/publications/dagger.pdf