So, as a justification for support of scala, the thing that seems lacking to me is that Chisel seems to still be centered around scala 2? E.g. their recommended template for getting started still uses scala 2 ... so without support to motivate them to use scala 3, it's not obvious that Chisel benefits much from current work on the scala language? I have not fully understood the Chisel project but I see they have a "compiler plugin" which suggests to me that moving fully off scala 2 requires a meaningful redesign.
Chisel absolutely isn’t the type of software that benefits from upgrading because it’s largely standalone. They could be the last project still stuck on 2.x a decade from now and it wouldn’t make much of a difference to its users.
I’ve only used Chisel for a few projects but I’ve never used anything but Chisel in those codebases. Simulation, verification, and all the painful stuff in FPGA/ASIC development depends on non-Scala tooling and all of the inputs (parametrization) are just read in from JSON files produced by scripts in other languages.
It would be nice to be up to date but the hardware NRE is so damn high that working around any limitations in Scala support is a rounding error. Chisel’s outputs are sent out for $X00,000 fab production runs so no one gives a damn whether it’s Scala 2 or 3 as long as it ships a working IC. The last time I used Chisel I was working on a mixed signal design where the Synopsys Fusion Compiler (maybe Custom compiler?) licenses alone ran into the hundreds of thousands per year (iirc it was per seat, so we must have spent over a million per year on Synopsys alone).
I’m not super plugged into scala but I work with Spark quite a bit and my observation has been the whole scala 2.13 -> 3 transition is a huge mess for just about everyone who touches it. I don’t have enough hands-on context to understand why it’s so painful but it seems to be similar or worse to the python 2.7 -> 3 transition in terms of sticking friction.
It is a mess. I've spent some time trying to convert some academic oss projects and some removed features really force large redesigns.
I think rather than funding the stuff on this announcement, I wish they would fund a team of experts to work on migration of a prioritized list of projects. This would both provide example patterns of migrating substantial projects and unblock projects who have been saying "we would like to try migrating but library X we use still hasn't"
Well, a more optimistic take here is that if future development on the Scala language was funded explicitly by/for people who are current using Scala 2, that means that the developers would more clearly understand their requirements in terms of making an easier transition for users moving from Scala 2 -> 3
> "Is money a birthright now? Do we just get born and get money from the government?" Republican Rep. John Gillette told Business Insider. "Because I think the Founding Fathers would say that is very contrary to our capitalist system and encouraging people to work."
I think the other perspective is that the US _has_ historically been comfortable with giving away property in the Homestead Act, land rushes, the Oregon Land Donation act etc. These were giving away _capital_, and the people needed to provide the labor to make it productive.
IDK what a future-facing equivalent should look like. Should we all own datacenter racks? Robot factory lines?
RAM. The pro-human state of the future has fabs & distributes ram to it's citizen.
I actually do think it would be incredibly interesting to have state sponsored data centers. Give folks a VPS. Give folks some incredibly affordable services.
Everyone gets their own Ai chat bot at birth (maybe with a physical robot for it to pilot) and it's up to them to direct it in ways that produce value?
- Most Americans don't know how social security is funded but
- An overwhelming majority of nonretired adults (79%) do not believe they will receive their full scheduled Social Security benefits when they retire.
- About three in four adults (77%) have heard that Social Security is projected to run short of money by 2033.
If you thought you had an individual account, and you also thought the program overall was going to run out of money and that you weren't going to get the "full benefit", does that mean you thought you weren't paying enough in?
Having an individual account does not mean that money earmarked for you is sitting in a vault. E.g, our banking system gives everyone an individual account, but does not have enough cash to pay everyone back; and occasionally their other assets fall, resulting in them not being able to satisfy their obligations to account holders. This has happened often enough, that bank accounts are not required to be insured against it.
There has also been a common talking point of Congress stealing from social security to fund other expenses, which could also explain how it would not be able to pay even with individual accounts.
How much should I put in isn't knowable - if I have known many who died before 62, and others who lived to 104. then there is various health that affects what you can spend - both ways, low health may mean spend nothing since you can't do anything or it may mean expensive nursing homes
I was just having a conversation (argument) with my MIL about this. She claims that people have felt that SS would run out of money before they retire her entire life. Has this sentiment been tracked at all? I certainly feel like I (aged mid-30s) will not be benefitting, but the common argument is that the Ponzi scheme will find a way to perpetuate for longer
Social Security counts on there being a larger workforce every year.
If there's ever a significant population crunch of working age people, it would be in trouble. In 2018, the cost of the Social Security program exceeded total revenues for the first time.
At current rates, Social Security is projected to be insolvent by 2034.
The only ways to mitigate this is to spend less or make more. But no one wants to pay more and no one wants to give up anything.
That early-mid 2030s prediction has been consistent all my life. It has gone up a little as rules changed but nobody was ever willing for the big reforms that could make it work long term.
Impossible to tell, but there's factors making it more likely to happen this time than in previous generations. Probably the biggest factor is the declining population, which no government around the world has been able to materialise an answer to.
My personal retirement plans assume no social security. Anything I may get in old age is a nice bonus, but in my planning, it's not the beef.
>which no government around the world has been able to materialise an answer to.
I disagree. The whole hot political topic of "illegal immigration" is as a side effect - seemingly a successful solution to motivate (foreign) people to bump the population. Specifically - "if you have a kid in the US, the kid is automatically a citizen, you don't have to pay taxes, you don't need a green card, you work under the table so make $0 on paper and thus qualify for free food and housing (welfare) and free healthcare (medicaid)".
Honestly if we gave such a deal to young non-foreign american women free and clear, likely we'd be seeing a lot more childbirths.
Despite all the economic burden and political strife this program has caused, it may in fact extend the life of the nation and welfare.
This is anecdotal, but I remember this sentiment becoming widespread in the 1980s, along with the imminent hyperinflation and currency collapse.
Political candidates (Carter and Reagan) promised reductions in federal spending, but couldn't get past the obvious lack of any real discretionary control due to the size of the biggest ticket items such as the entitlements and military. The conservative message of "entitlements will make you immoral" was replaced by "entitlements are unsustainable," making it sound like serious economics but without any real basis.
I think the "last 100 years with no maintenance" is not likely feasible with this approach. The top plate has a coating that supports high infrared emissivity -- and I think it would need to be regularly cleaned to work well. And you can't really prevent it from getting dirty by enclosing it b/c that both substantially changes the performance and moves the maintenance burden to cleaning the enclosure.
> Don't expect universities to sit still and do nothing.
> The popularity of residential degree programs may be ending, due to insanely high cost.
I think the problem is that universities _have_ been changing in the direction of _delivering less_ at the same time that they cost more. The article cites public schools doubling tuition in inflation-adjusted terms since 1995, but simultaneously:
- student-faulty ratios have gotten worse
- schools use under-paid adjuncts for a larger share of classes
- good schools often trade on the research record of faculty, but the success of those prominent faculty often mean they can get course buyouts / releases, so they're not teaching anyway
Rather than trying to make new online offerings, I think schools need to lean out their staff, and cut back on programs that don't have to do with instruction. Even better would be if federal funding eligibility was tied to schools demonstrating that at least X% of their budget goes to instruction, where that X should ratchet up over time.
The author neglects to observe that doubling tuition over 30 years equates to only a 2.35% inflation rate. That sounds pretty close to the US inflation rate during that time, so increases in tuition cost have been held in check pretty well.
No, if anything the article has a copyedit error in saying twice in one sentence that the doubling is after inflation-adjustment.
> While there have been some small declines in tuition prices over the last decade, when adjusted for inflation, College Board data shows that the average, inflation-adjusted cost of public four-year college tuition for in-state students has doubled since 1995.
Dedicated grad schools that are separate from, but affiliated with, dedicated undergrad schools. Those teaching at the dedicated undergrad schools will be hired for their ability to focus on foundational teaching, with research programs designed to involve undergraduate student researchers in genuine research, while still providing publication opportunities and genuine advancement of the art.
The talk focuses for a bit on having pure data from before the given date. But it doesn't consider that the data available from before that time may be subject to strong selection bias, based on what's interesting to people doing scholarship or archival work after that date. E.g. have we disproportionately digitized the notes/letters/journals of figures whose ideas have gained traction after their death?
The article makes a comparison to financial backtesting. If you form a dataset of historical prices of stocks which are _currently_ in the S&P500, even if you only use price data before time t, models trained against your data will expect that prices go up and companies never die, because they've only seen the price history of successful firms.
Not a financial person by any means, but doesn't the Black Swan Theory basically disproves such methods due to rarity of an event that might have huge impact without something to predict (in the past) that it might happen, or even if it can be predicted - the impact cannot?
For example: Chernobyl, COVID, 2008 financial crisis and even 9/11
I think therapists in training, or people providing crisis intervention support, can train/practice using LLMs acting as patients going through various kinds of issues.
But people who need help should probably talk to real people.
Yeah, I think the crappy side of it at this point is that the biometric data they collect is never leveraged to help you as a citizen.
If I lose my passport while abroad, given that the government has my fingerprints etc, why can't I use those biometrics to reenter the country (and have a replacement passport reissued immediately)?
Officially, you are supposed to be able to opt out of the face recognition cameras at security but I think whether staff actually respect that is not consistent.
If you strongly feel that AI is doing analysis and decision-making at a level which is comparable to human executives, you and some like-minded people could try to form a PE group that takes over firms, replaces a highly compensated CEO with a lower compensation one and an OpenAI subscription, and see how you do.
However, only months ago Anthropic published about their autonomous vending machine experiment which had both humorous and interesting but not financially successful. I think maybe we're not to the point where AIs can run businesses. https://www.anthropic.com/research/project-vend-1
The skill ladder for humans assumes CEO functions require higher and better skills than a vending machine manager. But there's no reason to assume that AI capabilities map to those jobs in the same way.
It's quite possible that an AI would do a better job at CEO than running a vending machine.
And it's quite possible that successful CEOs would not succeed at managing a vending machine. It's almost guaranteed that many will fail, given that some C-level people are selected for risk appetite and social skills over actual business acumen.
I encourage you to read the Anthropic blog post, which includes situations like:
- hallucinating conversations with fictitious people
- role playing as a real person and telling them they are present in meatspace, or claimed to have visited a fictional location
- hallucinating details of an account involved in payments
- both ignoring lucrative opportunities and accepting loss-making deals
- making commitments before having done any research
While yes, the skills to run a vending machine and lead a company are not exactly the same, I think the nature of failures discussed means they would likely affect both roles.
I think there's every possibility that a present-day AI allowed to act as CEO would make mediocre choices for some period and then decide that it was a character in a scifi novel selling bespoke space-yachts to comet-mining magnates.
I guess you could make the argument that we haven't proven AI can't do a CEO's job until we test it... it would make an interesting experiment, at least.
But it also kind of reminds me of the DAO hype a few years ago where people suggested decentralized, blockchain-based organizations could be the org structure of the future. (Not realizing that they were just putting a co-op on a blockchain :) )
Now that I think about it, I do hope someone tests this theory out. It would be very interesting to see. You could probably start by saying "You are the CEO of a lemonade stand. Your goal is to maximize revenue while [constraints]..."
It's entirely possible... In fact, I'm pretty sure that there are more than a few very wealthy people willing to place bets on that very thing... maybe not everything, but upwards of 1-3% of their wealth.
Now, actually getting in a room and talking to those people and getting in on the execution of that idea... that's a different story.
To tie this specific example to a larger framework: In scala land, Tiark Rompf's Lightweight Modular Staging system handled this class of metaprogramming elegantly, and the 'modular' part included support of multiple compilation targets. The idea was that one could incrementally define/extend DSLs that produce an IR, optimizations in that IR, and code generation for chunks of DSLs. Distinctions about stage are straight-forward type-signature changes. The worked example in this post is very similar to one of the tutorials for that system: https://scala-lms.github.io/tutorials/regex.html
https://github.com/chipsalliance/chisel-template/blob/main/b...
reply