throwaway0123_5

> high income from software jobs

I don't think it is a worry about just software jobs though. Would people be nearly as concerned if some "Technology X" would delete all software jobs but leave all other white-collar jobs intact? Probably not nearly as much, just retrain and go do some other job. I think the concern is that once an AI can replace SWE, it is likely already at the level where in short order there will be no value to intellectual labor, no "thinking jobs" that can sustain a good life.

So basically I don't think it is just self-interest. I think a lot of people see a plausible outcome being the destruction of the value of (at least intellectual) labor. If labor loses all value and society isn't prepared to entirely upend our economic system, most people could suffer greatly, to the extent where the positive and exciting benefits of AI don't really mean much. If someone believes that not making their software open-source can delay that, it isn't necessarily a selfish decision. If that is what you believe, you're delaying a negative outcome for a lot of people.

> I would suggest you question and analyze your motivations.

Full disclosure up front: I'm not anti-AI and think there are a lot of potential positive applications. I currently work on improving some aspects of AI.

That said, it really isn't that hard to see why many people (including me to some extent) are nervous about AI getting better, and upset that the people who made the content that facilitated this are not benefiting from it.

If you're a SWE and not already extremely stable financially, AI replacing software engineering (and if that happens, ~all white-collar labor) means that you'll likely be unable to achieve anything close to the economic status expected and planned on.

> Before LLM, anyone could already take free software and copy-paste the code while stripping away the author's name or the license.

Before LLM, people doing this posed ~0 existential threat to the livelihood of the author. AI doing it at a massive scale arguably does (remains to be seen, but at least plausible).

I think most people want their work to have a positive impact, and aren't that concerned about credit. But they aren't confident that helping LLMs will lead to long-term positive impacts, and in any case are less worried about that than not being able to own a house or maybe even buy necessities because of the possible economic impacts of AI.

Is this more of an indictment of our current social/economic system than of AI itself? Probably yes, but I'm not sure it matters.

I think it would be an alright (potentially good) outcome if in the short-term we don't see major progress towards AGI.

There are a lot of positive things we can do with current model abilities, especially as we make them cheaper, but they aren't at the point where they will be truly destructive (people using them to make bioweapons or employers using them to cause widespread unemployment across industries, or the far more speculative ASI takeover).

It gives society a bit of time to catch up and move in a direction where we can better avoid or mitigate the negative consequences.

I'm curious why there are no results for the "Claude 3.7 Extended Thinking" on SWE-Bench and Agentic tool use.

Are you finding that extended thinking helps a lot when the whole problem can be posed in the prompt, but that it isn't a major benefit for agentic tasks?

It would be a bit surprising, but it would also mirror my experiences, and the benchmarks which show Claude 3.5 being better at agentic tasks and SWE tasks than all other models, despite not being a reasoning model.

> For django-32517

Although I agree with your analysis and it doesn't look great for the authors, this issue (https://code.djangoproject.com/ticket/32517) arguably falls into their "Solution leak" category anyways, as the following text appears in the issue description (and so I think directly in `problem_statement` rather than `hints_text`):

> Currently, OrderedSet isn't reversible (i.e. allowed to be passed as an argument to Python's reversed()). This would be natural to support given that OrderedSet is ordered. This should be straightforward to add by adding a __reversed__() method to OrderedSet.

It isn't the exact code though, so I suppose it could be argued instead that the issue is just extremely easy.

As a sibling comment implies though, there's also danger from it being stupid while unsupervised. For example, I'd be very nervous having it do something autonomously in my kitchen for fear of it burning down my house by accident.

This is a pretty big range. Are you saying the overall multiplier for all of your work is somewhere between 2 and 10x, or that some tasks are 2x and others are 10x?

Although I wouldn't bet against o3, I think it works to their favor to release it later no matter how well it is doing.

Case 1, does worse than or is on-par with o1: Would be shocking and not a great sign for their test-time compute approach, at least in this domain. Obviously they would not want to release results.

Case 2, slightly better than o1: I think "holding off until they have a chance to tune it further" applies.

Case 3, does much better than o3: They get to release it after another model makes a noticeable improvement on the benchmark, get another good press release to keep hype high, and they get to tune it further before releasing results.

They do address the potential for contamination in the paper fwiw:

> Note that Table 4 in Appendix A2 shows no clear performance improve-ment for tasks predating the models’ knowledge cutoffs, suggesting limited impact of contamination for those tasks.

> yet o1 with high reasoning could only solve 16.5% of tasks formatted this way.

48.5% with pass@7 though, and presumably o3 would do better... they don't report the inference costs but I'd be shocked if they weren't substantially less than the payouts. I think it is pretty clear that there is real economic value here, and it does make me nervous for the future of the profession, moreso than any prior benchmark.

I agree it isn't perfect. Only tests TS/JS and the vast majority of the tasks are front-end, still none of the mainstream software engineering benchmarks test anything but JS/Python/sometimes Java.

> Turns out, it's worse than Claude Sonnet when it comes to coding.

This was an interesting takeaway for me too. At first I thought that it suggested reasoning models mostly only help with small-scale, well-defined reasoning tasks, but they report o1's pass@1 going from 9.3% at low reasoning effort to 16.5% with high reasoning effort, so I don't think that can be the case.

Tech workers got fired so FBI/IRS workers should also get fired? What is the logic here? As far as I'm aware the FBI/IRS have nothing whatsoever to do with the tech job market, let alone individual employees.

As far as public sentiment wrt the firings go, it isn't super surprising that people are more upset about government employees getting fired. Big tech is becoming increasingly associated with things people don't like. Ads, toxic social media, gig economy, annoying CEOs (admittedly not the fault of the workers), AI in many cases, scams, pricing people out of cities they grew up in, etc. I think if you randomly selected a federal government employee and a tech worker, there's a much better chance the government employee is doing something more helpful for the average American.

It seems like there was a dip in mid-May and that it didn't go back up until mid/late-August? That corresponds pretty closely to US summer breaks. Also a huge spike in early December followed up a huge drop, final exams and then winter break?

I agree with this. An AI that can fully handle web dev is clearly AGI. Maybe the first AGI can't fully handle OS kernel development, just as many humans can't. But if/once AGI is achieved it seems highly unlikely to me that it will stay at the "can do web dev but can't do OS kernel dev" level for very long.

Likewise, under no circumstances would I trade C for LLM-aided assembly programming. That sounds hellish. Of course it could (probably will?) be the case that this may change at some point. Innovations in higher-level languages aren't returning productivity improvements at anywhere close to the same rate as LLMs are, and in any case LLMs probably benefit from improvements to higher-level languages as well.

Some codebases grown with AI assistance must be getting pretty large now, I think an interesting metric to track would be percent of code that is AI generated over time. Still isn't a perfect proxy for how much work the AI is replacing though, because of course it isn't the case that all lines of code would take the same amount of time to write by hand.

In all seriousness, what was his point?

Because he says:

> If he’s a bad dude or a terrible member of the team, fire him for that.

Saying things like "Normalize Indian hate" and "I was racist before it was cool" are statements that pretty well support a "bad dude" classification, especially in the context of a role where he'll be responsible for firing people and removing "DEI hires." Even if you agree (with Musk, Elez, etc.) that DEI is a problem, hiring a self-avowed racist to deal with it seems pretty idiotic.

Calling him "a kid" is also pretty disingenuous. He was 24 or 25 when he made those posts, hardly a kid. And they weren't ancient history, they were last year. I'm also for forgiveness if time has passed and people show that they deserve it, but it isn't ruining his life to hold him accountable to things he said last year, as an adult.

> Nor can an overly permissive reading be justified on moral grounds... most of Europe (and indeed, the world) does not honor the concept of jus soli.

It is extremely common in the Americas though. I think only Colombia and a few island countries don't have birthright citizenship here. I think it is a good concept for us, the US has historically been a nation of immigrants and our country has a culture that is shaped (and IMO strengthened) by people from all over the world.

I'm not completely convinced of this, even in the presence of AGI that is peak-human intelligence in all ways (lets say on-par with the top 1% researchers from top AGI labs, with agency and online learning are fully solved). One reason for this is what the sibling comment argues:

> Exponentially smarter AI meets exponentially more difficult wins.

Another is that it doesn't seem like intelligence is the main/only bottleneck to producing better AIs right now. OpenAI seems to think building a $100-500B data center is necessary to stay ahead*, and it seems like most progress thus far has been from scaling compute (not to trivialize architectures and systems optimizations that make that possible). But if GPT-N decides that GPT-N+1 needs another OOM increase in compute, it seems like progress will mostly be limited by how fast increasingly enormous data centers and power plants can be built.

That said, if smart-human-level AGI is reached, I don't think it needs to be exponentially improving to change almost everything. I think AGI is possibly (probably?) in the near-future, also believing that it won't improve exponentially doesn't ease my anxiety about potential bad outcomes.

*Though admittedly DeepSeek _may_ have proven this wrong. Some people seem to think their stated training budget is misleading and/or that they trained on OpenAI outputs (though I'm not sure how this would work for the o models given that they don't provide their thinking trace). I'd be nervous if it was my money going towards Stargate right now.

> Some very simple explanations, the models were probably overfitted to a degree on Python given its popularity in AI/ML work, and SWE-bench is all Python. Also, the underlying Github issues are quite old, so they probably contaminated the training data and the models have simply memorized the answers.

My understanding is that it works by checking if the proposed solution passes test-cases included in the original (human) PR. This seems to present some problems too, because there are surely ways to write code that passes the tests but would fail human review for one reason or another. It would be interesting to not only see the pass rate but also the rate at which the proposed solutions are preferred to the original ones (preferably evaluated by a human but even an LLM comparing the two solutions would be interesting).

I've seen this figure on a few tech news websites and reddit but can't find an official source. If it was in the video I must have missed it, where is this coming from?

Life expectancy and median age (didn't find average but I assume it is pretty close) were the first things I checked when I read your comment, but life expectancy globally is in the 70s and the median person about 30. I don't understand how the average person could have only a little more than 10 years left.

Are you suggesting a massive collapse is likely in ~10 years?

Yeah that certainly seems hyperbolic. My opinion is that we need to do better in the future and consider our past when making current policy decisions. And in some cases I think we should proactively attempt to make amends.

I'm not a web dev at all, but when I need some web UI ClojureScript/Reagent works pretty well. I get all the pretty React components while being able to avoid JavaScript and JSX entirely (all XML-style DSLs that have closing tags I find incredibly verbose and ugly).

Exactly... I find all the arguments in the style of "but we need food" extremely disingenuous when it comes to meat production. Almost without exception, more food could be produced by converting land used for animal agriculture into land used to grow food directly for human consumption.

Tbh it seems entirely plausible to me that a messenger being unable to deliver an important message could have an outsized effect on that outcome of a battle. What if they're letting their side know about a surprise attack?

They're talking about a horseshoe on a horse which was being used to deliver an important message

> That's an argument that the US isn't morally perfect.

It certainly seems to me like its also a strong argument that much of the US's wealth is based on colonialism or colonialism-adjacent policies, no?

> I don't see how this invalidates the idea that the US culture is better at creating and running a great economy

I didn't argue that. If US culture is/was in favor of colonialist antics and colonialism produces wealth for the US, that would actually be an argument in favor of US culture being better at producing a great economy. I would argue that the ends don't justify the means when the means are abusing far poorer neighboring countries.

I also wouldn't argue that colonialism is the only reason the US is wealthy. There are clearly aspects of US culture that are conducive to productivity and innovation that are more or less independent of colonialism.

Colonial-esque behavior by the US was (is?) hardly limited to plantation slavery in the US south. For much of the late 1800s and most of the 1900s the US government was more than happy to intervene in the domestic affairs of other countries to protect corporate profits. One particularly egregious example is the CIA-aided overthrow of Jacobo Arbenz in Guatemala in the 50s, largely to protect the profits of US fruit companies, but you don't have to look far to find more.

From General Smedley Butler, most decorated marine at the time of his death and the only marine with two medals of honor:

> I spent 33 years and four months in active military service and during that period I spent most of my time as a high class muscle man for Big Business, for Wall Street and the bankers. In short, I was a racketeer; a gangster for capitalism. I helped make Mexico and especially Tampico safe for American oil interests in 1914. I helped make Haiti and Cuba a decent place for the National City Bank boys to collect revenues in. I helped in the raping of half a dozen Central American republics for the benefit of Wall Street. I helped purify Nicaragua for the International Banking House of Brown Brothers in 1902–1912. I brought light to the Dominican Republic for the American sugar interests in 1916. I helped make Honduras right for the American fruit companies in 1903. In China in 1927 I helped see to it that Standard Oil went on its way unmolested. Looking back on it, I might have given Al Capone a few hints. The best he could do was to operate his racket in three districts. I operated on three continents.

> The vast majority of the population choose where the want to live first – in fact, the majority of the population still live within a small radius of where they were born!

I wouldn't call it a vast majority, apparently 59% of people live in the state where they were born and most states are pretty big places (https://www.test.census.gov/newsroom/blogs/random-samplings/...).

> Moving somewhere for a career is fairly abnormal.

I think you're probably right on a job-to-job basis, most people aren't picking up and moving across the country for each new role. But it only takes one move to another city for the "several multi-generational families in close proximity" dynamic of many rural areas to be disrupted. Even short moves could easily make someone much more socially isolated. Move 50 miles away from your hometown and now you're seeing your former neighbors once a month or less instead of a few times a week.

> there is a small segment of the population who will chase work at the cost of where they live. Let's say this is who ends up in the suburbs. Perhaps that's the problem? As in they end up being comprised of people focused on their career, and thus don't prioritize community?

Not sure this is the right way to frame it. It isn't necessarily about "ending up" in the suburbs when a huge percentage of the country was born in the suburbs or in a city, never having had a tight-knit multigenerational community to begin with. 80% of the US is urbanized (https://en.wikipedia.org/wiki/Urbanization_in_the_United_Sta...). Most people are moving from generic suburb to generic suburb (or city). A kid born in some suburb isn't choosing to focus on their career over community, but nevertheless economic migration was likely the force that caused them to end up there.

> Does that actually appeal to the people of the suburbs, or are you projecting?

Somewhat remains to be seen, but my gut feeling is yes. I think that most suburbanites haven't deeply considered other alternatives given that they've mostly only been exposed to "default US suburbia." That was the case for me until I got into urbanist YouTube and moved to a more urbanist location. When I share urbanist material with friends and family that haven't been exposed to it before, they tend to be pretty receptive. Anecdotes yes, but I'm not being disingenuous.

> constituents of a democratic government, and therefore can already have anything their collective hearts desire. Why isn’t this already the reality?

In theory yes, but in reality these are changes that will take a long time. There's a lot of red tape when it comes to building and zoning, and vocal minorities (NIMBYs) can often block or delay efforts that have popular support through lawsuits (recently near my area: https://www.arlnow.com/2024/09/27/breaking-judge-overturns-m...). Consider that there are plenty of issues that have wide bipartisan support among US voters as a whole but haven't been implemented for political reasons. Even when there is popular support and policy is implemented, whole areas can't simply be rebuilt overnight.