That's why in my workflow I don't write single monster specs. Rather, I work with the LLM to iterate on small, individual, highly constrained specs that provide useful context for what/why/how -- stories, if you will -- that include a small set of critical requirements and related context -- the criteria by which you might "accept" the work -- and then I build up a queue of those "stories" that form a, you might say, backlog of work that I then iterate with the LLM to implement.
I then organize that backlog so that I can front-load uncovering unknowns while delivering high-value features first.
This isn't rocket science.
By far the biggest challenge I experience is compounding error during those iterative cycles creating brittleness, code duplication, and generally bad architecture/design. Finding ways to incorporate key context or other hints in those individual work items is something I'm still sorting out.
As a homeowner I literally pay a fee every month to fund the construction and maintenance of the infrastructure that gets water to my home.
If that water supply is cut off without valid reason, there is a complaint mechanism with the local utility commission where the issue can be heard and resolved.
I also live in the municipality where my water is supplied, and therefore am represented by its government.
Therefore I am both economically and politically invested in that infrastructure.
Unfortunately none of this is true for the folks in this article.
Is that a "guarantee"? No. Nothing is guaranteed. But it's far better than the arrangement these folks operated under.
Yup, fair. I tried to acknowledge that in my paragraph about KYC in a follow-up edit to one of my earlier comments, but I agree, the language I've been using has been intentionally quite strong, and sometimes misleadingly so (I tend to communicate using strong contrasts between opposites as a way to ensure clarity in my arguments, but reality inevitably lands somewhere in the middle).
#1. Facebook did everything they could to evaluate Flo as a company and the data they were receiving, but they simply had no way to tell that the data was illegally acquired and privacy-invading.
#2. Facebook had inadequate mechanisms for evaluating their partners, and that while they could have caught this problem they failed to do so, and therefore Facebook was negligent.
#3. Facebook turned a blind eye to clear red flags that should've caused them to investigate further, and Facebook was malicious.
Personally, given Facebook's past extremely egregious behaviour, I think it's most likely to be a combination of #2 and #3: inadequate mechanisms to evaluate data partners, and conveniently ignoring signals that the data was ill-gotten, and that Facebook is in fact negligent if not malicious. In either case Facebook should be held liable.
pc86 is taking the position that the issue is #1: that Facebook did everything they could, and still, the bad data made it through because it's impossible to build a system to catch this sort of thing.
If that's true, then my argument is that the system Facebook built is too easily abused and should be torn down or significantly modified/curtailed as it cannot be operated safely, and that Facebook should still be held liable for building and operating a harmful technology that they could not adequately govern.
If Facebook chooses to build a system that can ingest massive amounts of third party data, and cannot simultaneously develop a system to vet that data to determine if it's been illegally acquired, then they shouldn't build that system.
You're running under the assumption that the technology must exist, and therefore we must live with the consequences. I don't accept that premise.
Edit: By the way, I'm presenting this as an all-or-nothing proposition, which is certainly unreasonable, and I recognize that. KYC rules in finance aren't a panacea. Financial crimes still happen even with them in place. But they represent a best effort, if imperfect, attempt to acknowledge and mitigate those risks, and based on what we've seen from tech companies over the last thirty years, I think it's reasonable to assume Facebook didn't attempt similar diligence, particularly given a jury trial found them guilty of misbehaviour.
> None of your example have anything to do with the thing we're talking about, and are just meant to inflame emotional opinions rather than engender rational discussion about this issue.
Not at all. I'm placing this specific example in the broader context of the tech industry failing to a) consider the consequences of their actions, and b) escaping accountability.
We're seeing this broad trend in tech where we just want to shrug and say "gee wiz, the machine did it all on its own, who could've guessed that would happen, it's not really our fault, right?"
LLMs sharing dangerous false information, ATS systems disqualifying women at higher rates than men, black people getting falsely flagged by facial recognition systems. The list goes on and on.
Humans built these systems. Humans are responsible for governing those systems and building adequate safeguards to ensure they're neither misused nor misbehave. Companies should not be allowed to tech-wash their irresponsible or illegal behaviour.
If Facebook did indeed built a data pipeline and targeting advertising system that could blindly accept and monetize illegally acquired without any human oversight, then Facebook should absolutely be held accountable for that negligence.
"Warning, this washing machine might burn your house down" is not sufficient to escape punishment. Why should digital technology get a pass just because the product that's offered is intangible?
Look back. At no point did I suggest AI should be banned or outlawed. My remedy for washing machines burning down houses isn't to ban washing machines. It's to ensure there are appropriate incentives in place (legal, financial, reputational) to encourage private industry to consider the potential negative externalities of what they're doing.
This is the story of the modern tech industry at large: a major new technology is released, harms are caused, but because of industry norms and a favourable legal environment, companies aren't held liable for those harms.
It's pretty amazing, really. Build a washing machine that burns houses down and the consequences are myriad and severe. But build a machine that allows countless people's private information to be leaked to bad actors and it's a year of credit monitoring and a mea culpa. Build a different machine that literally tells people to poison themselves and, not only are there no consequences, you find folks celebrating that the rules aren't getting in the way.
> It's a logical presumption. Researchers discover things. AGI is a researcher that can be scaled, research faster, and requires no downtime.
Those observations only lead to scaling research linearly, not exponentially.
Assuming a given discovery requires X units of effort, simply adding more time and more capacity just means we increase the slope of the line.
Exponential progress requires accelerating the rate of acceleration of scientific discovery, and for all we know that's fundamentally limited by computing capacity, energy requirements, or good ol' fundamental physics.
Consider theoretical physics, which hasn't significantly advancement since the advent of general relativity and quantum theory.
Or neurology, where we continue to have only the most basic understanding of how the human mind actually works (let alone the origin of consciousness).
Heck, let's look at good ol' Moore's Law, which started off exponential but has slowed down dramatically.
It's said that an S curve always starts out looking exponential, and I'd argue in all of those cases we're seeing exactly that. There's no reason to assume technological progress in general, whether via human or artificial intelligence, is necessarily any different.
The parent is referring to the "buy, borrow, die" strategy of wealth accumulation. Would that work in your parent's specific circumstance? Maybe? Maybe not? But taking a low interest loan against assets as a method of wealth generation and tax avoidance is both a viable strategy and an extremely popular one.
> There are many ways to make consumption taxes not regressive. You can implement refunds up to certain threshold. You can use the revenue to fund services used by lower income families
Every one of those "solutions" is just a patch on the basic problem that consumption taxes are fundamentally regressive.
Go tell someone living paycheque to paycheque that it's okay, you'll get a rebate every quarter for the extra tax they paid, or worse, on their annual tax filing, and tell me how that'll make their household budget actually work.
Honestly, when I read ideas like this, I realize just how massive the disconnect is between the lived experiences of the relatively well off and the working poor...
> And given we lack the technical ability to fully understand global warming
I'm sorry, what?
We have total understanding of global warming: 1) CO2 traps heat, we know that from basic lab experiments, and 2) we're pumping massive amounts of CO2 into the atmosphere, which we know from countless data points, both historical and contemporary.
When it comes to the basic fact of global warming, that's all you need to know.
Everything after that is forecasting impacts and associated trend lines.
Yeah, LLMs share a lot of the same challenges as self-driving cars: when they work great, we get complacent, and then when they fail, they fail in ways that humans are really bad at anticipating.
And the discourse around all this is also the same: the detractors point out these flaws, and the proponents chime in with "Yeah well humans suck too" and around and around we go...
> Conservative ideology views government research as inefficient compared to private sector research, something I generally agree with.
I'd agree that's generally true when a profit motive is identifiable.
Absent that, if left to the private sector, the research simply doesn't get done at all.
> The evidence for this in regards to space is very clear IMO
You have an incredibly narrow view of what space "research" involves. It's not just about putting payloads in LEO. Were it up to the private sector, the HST wouldn't exist at all.
Whether that's good or bad is a philosophical discussion, but personally, I believe science should be about more than just what is profitable.
(and that's ignoring that, as others have pointed out, SpaceX stands on the shoulders of government-funded giants)
The reality is there's room for both government-funded and private research, and to imply otherwise... well, let's just say we're living in a time of extremely narrow, black-and-white, polarized thinking, and this is an excellent example.
It says a lot about the current discourse around AI that 6 years ago Marcus would write:
> Despite all of the problems I have sketched, I don’t think that we need to abandon deep learning.
And that would somehow be spun, today, as "LLMs are the wrong approach".
Meanwhile, another attempt to post this article here got straight up flagged, I can only assume because this whole topic has become about religious orthodoxy vs the heretics.
That's why in my workflow I don't write single monster specs. Rather, I work with the LLM to iterate on small, individual, highly constrained specs that provide useful context for what/why/how -- stories, if you will -- that include a small set of critical requirements and related context -- the criteria by which you might "accept" the work -- and then I build up a queue of those "stories" that form a, you might say, backlog of work that I then iterate with the LLM to implement.
I then organize that backlog so that I can front-load uncovering unknowns while delivering high-value features first.
This isn't rocket science.
By far the biggest challenge I experience is compounding error during those iterative cycles creating brittleness, code duplication, and generally bad architecture/design. Finding ways to incorporate key context or other hints in those individual work items is something I'm still sorting out.
(and yes, I use en-dashes, and no I'm not an AI)