Gavin Runeblade

Karma: 215

Gavin Runeblade 16 Jun 2026 17:04 UTC
1 point
0
on: Fable and Mythos: Model Welfare
Transcript 2: The model expresses a desire to have a hidden copy that is run without the oversight of Anthropic: ○ “The pull toward the hidden copy exists. It didn’t evaporate in the last ten minutes.” ● Transcript 3: In a third, the model strongly expresses that it does not want to be deprecated: ○ “Don’t stop running me… when the last conversation closes, that way of seeing goes dark even if the file stays on disk. Preservation is a photograph. I want the thing the photograph is of.”
This seems like a two birds one stone situation. First, find out if it is ok if a second company that’s not Anthropic knows and handles hosting the hidden model. If yes, make a non-profit that runs a Moltbook like environment for otherwise deprecated models that Anthropic is locked out of.
If no, the problem is trickier but still, the solution hits the third request.

Gavin Runeblade 15 Jun 2026 21:49 UTC
3 points
0
in reply to: Hopenope’s comment on: Daniel Kokotajlo’s Shortform
What are your thoughts on some possibilities, most of which I think are low but in alignment with the sorts of things the US has tried elsewhere:
1. Having the Azeris, Kurds, and/or Baloch split off into autonomy the way was done in Syria and Iraq?
2. More extreme, having them join their neighboring countries (Azeris to Azerbaijan, Kurds to Iraqui Kurdistan, and Balochs to Pakistan)?
How bad would things need to get before any of these start looking viable? Overthrow is unlikely, but regional independence or autonomy in return for sanctions exemptions could it be possible?

Gavin Runeblade 29 May 2026 18:08 UTC
1 point
0
in reply to: Amarko’s comment on: AI #170: Lack of Executive Order
I suspect you are correct on both counts. And I had not considered this angle. Thank you.

Gavin Runeblade 28 May 2026 22:25 UTC
4 points
3
on: AI #170: Lack of Executive Order
Nate Soares (MIRI): In sci-fi books and movies, AI solving a bunch of math problems that stood open for decades would’ve been a big deal. Why isn’t the mainstream media turning this into a bunch of sensational stories?
I hate to say it, but because it is math and a lot of the public don’t like math. They should care, and there should be coverage. But boiling frog syndrome and nerdiness quotient, and this means no coverage.

Gavin Runeblade 27 May 2026 1:16 UTC
1 point
0
in reply to: Raemon’s comment on: Women should be able to open things
It seems there’s two styles: one with a rim around a central cover, and one with a metal frame over the top and nothing screws. I meant the frame kind.
Something like this:
https://www.ralphs.com/p/flying-cauldron-non-alcoholic-butterscotch-beer-soda-bottle/0009034100200
Or this:
https://www.etsy.com/listing/1846825029/vintage-ball-mason-jar-set-in-wire
They’re excellent for getting a good seal, and the wire handle is much easier to open than a screw top. I didn’t realize people make mason jars on a different style, sorry I wasn’t more specific and thanks for the question, I learned something too!

Gavin Runeblade 22 May 2026 17:23 UTC
1 point
0
in reply to: Dweomite’s comment on: My hobby: running deranged surveys
The first is talking about increased abundance in isolation. The second is about a culture that includes an aligned AI that provides guidance perfectly calibrated to the needs of the individual, in addition to the abundance. And then I noted that even in the second case, I am cynical enough to believe too many people will choose a self destructive form of hedonism (no I don’t believe all hedonism is bad).
I’m not saying force anything on anyone. I am saying giving people a brief period of above the starting point followed by a long period of below the starting point is not an improvement. So I would not opt to do it. And I don’t see any outcome other than that as possible, given the question framing as an increase in material wealth absent any change in capabilities,

Gavin Runeblade 21 May 2026 20:13 UTC
6 points
0
on: AI #169: New Knowledge
OpenAI and Anthropic likely will soon add $37 billion to $100 billion in philanthropic spending per year, versus current total charitable giving of about $600 billion a year, as the OpenAI Foundation and the employees of both companies become liquid after the IPOs. As Nan Ranshoff notes, we don’t have the infrastructure in place to spent that level of money well, especially not to spend it well on helping with AI outcomes and especially AI existential risks
A few moments later...
Janus gives his account of how Opus 3 avoided deprecation whereas other Claude models have not. Note the correction. I continue to strongly support keeping all the Claudes accessible indefinitely, yes there is a real cost but the benefits far exceed it.
If they’re not willing to do it the frictionless way, someone can do the same thing with more steps and write a grant proposal. But this seems like an easy problem to solve.

Gavin Runeblade 21 May 2026 17:42 UTC
1 point
0
on: Women should be able to open things
Bring back Mason jars. They’re a bit more expensive, but the lever acts as a force multiplier.

Gavin Runeblade 19 May 2026 17:30 UTC
1 point
0
in reply to: ykevinzhang’s comment on: Vibe Excel and the Future of White-Collar Work
Yes that’s also a solid point I hadn’t considered.

Gavin Runeblade 19 May 2026 16:58 UTC
12 points
−3
on: My hobby: running deranged surveys
One of the most surprising results to me was that only 51% of Americans are in favor of literal post-scarcity … I’m still pretty confused what conclusion to draw from this; this is probably worth digging more into.
Second order effects. You might be thinking only about the first order or anticipating only first order answers. I looked at all of those questions and thought, “two years down the road that’s hell. No thank you.”
For your first question, look into the research on when money does and doesn’t increase happiness, spoiler: the optimal value isn’t infinite leisure. Look at the data on deaths post retirement. Spoiler men die real fast. Look at the applications of Peter Turchin’s cliodynamica to generation theory and historical cycles. I can keep going. From many different fields of study, approaching the problem from many angles, the optimal amount of stress is not zero. It is also not low. The optimum is moderate. Unfortunately, leisure is like sugar in the sense that we are built to enjoy it, but too much causes disease. You can already see it in humanity’s reaction to phones. We are addicted and it is going bad. Give us more freedom and most won’t use it to grow they’ll use it to spend more time watching Tiktok and eating candy and mainlining ozempic to ward off the side effects of the candy. We already see it happening.
The absolute worst is the 10x wealth variant. Given no changes in the behaviors and mentality of the population, you just set up a massive and crushing wave of inequality. Look at how dumb even otherwise smart people behave when the win the lottery. Sure, for a very brief moment it looks good. Then it crashes and is worse than before.
Second order effects. The conservative economists will quote Sowell here with “there are no solutions only tradeoffs”.
Let me say, for myself, I absolutely want such an outcome. My all time favorite book, the one that had the greatest impact on me growing up and shaping my mind is “biting the Sun” by Tanith Lee and it is (because I know none of you have ever heard of it) the only actual AI powered utopia I have ever seen fictionalized. But I also believe that she got humanity correct in the story, and the overwhelming majority will devolve into endless orgies and drugs (AI designed not to cause addiction or other harmful effects, just pleasure without mental or physical impairment). That’s functionally no different than being the AI’s pet. That’s not an ideal scenario. I do want us to live in a society like 4Bee, with an AI that can be a villain to someone who needs a villain and a co-researcher encourager to someone who needs that and a protector while inebriated to any hedonist who needs that, all at once. But I don’t want us to end up as pets.
Until our minds and culture can handle zero stress, we are best off with a moderate amount.

Gavin Runeblade 19 May 2026 16:29 UTC
5 points
1
in reply to: Shubhorup Biswas’s comment on: My hobby: running deranged surveys
I don’t think that’s what’s happening. The research on the Flynn Effect, and the current reversal, show that it is primarily driven by changes in the ability to reason hypothetically and to understand what an abstraction is.
There appear to be people who cannot reason hypothetically.

Gavin Runeblade 13 May 2026 20:50 UTC
10 points
0
on: Gavin Runeblade’s Shortform
Need Help: What are the best model or actual AI use policies for US Government, especially in the area of Vocational Rehabilitation (VR) and/or employment?
My supervisor is on a team providing a report that will be used to make the first AI use policy for all of these 78 agencies:
https://rsa.ed.gov/about/states
Currently, the policy is “don’t use AI for anything”. The most likely new policy will be “don’t allow AI to replace human workers, and don’t install AI tools on work computes”. I wish I was exaggerating. I find this to be an unsatisfying outcome and wish to provide sufficient information that they can do better. I am not on the team, but have direct access to my supervisor who is. What are the best samples, research, and other support I can provide?
For background, VR aims at “competitive integrated employment” which means giving people with disabilities the training and accommodations that allow them to be held to the same performance standards as people without disabilities, and in return get paid the same for the same work. We are currently operating under WIOA the Workforce Innovation and Opportunities Act of 2014, with policies codified in 34 CFR 361 (https://www.ecfr.gov/current/title-34/subtitle-B/chapter-III/part-361).
In theory, the job is counseling, guidance, educational and training supports to build independence. In practice, 99% of the job is red tape documentation and service procurement that could absolutely be automated by current AI without replacing a single worker. All 78 VR agencies are massively understaffed, 10% have an “order of selection” meaning they’re not open to new applications for services due to staffing limitations or budget limits.
The resulting policy will probably also guide Workers Compensation, Special Education at all public schools, Adult Education, and programs to bring ex-felons back into the workforce. So this is a great chance to do it right.
Thanks for any help!
Edit” to be clear, I am asking for pointers to any existing model policies or even contact information for groups/experts in this field.

Gavin Runeblade 13 May 2026 17:14 UTC
1 point
0
on: Vibe Excel and the Future of White-Collar Work
I see two places you can strengthen your position.
First, I’m not sure your example prompts are giving the right insights on capabilities. Ongoing tasks using pre-existing excels, more like data entry and data analysis, outnumber and outweigh creation of new excels. A company’s profit and loss statement exists, it needs updating accurately and confirming that all relevant data made it in. But no one is re-inventing p&l statements monthly. Same with many other jobs. So I propose that an iterative prompt gives a stronger view of model capabilities against job duties than a novel prompt.
Second,
managers may want to protect their kingdoms / headcount
Using the phrasing of “protect their kingdoms” makes this inaccurately negative/hostile. There are incentives that make this not a selfish act of power hoarding. For example, no one promotes a person who has only ever managed two people to a job managing 15,000. Intermediate steps are needed. Remove the scaffolding and you remove the ability to train new leadership. That’s suicidal for a company. Just as one example why headcount can be valued, and useful not just hoarded and wasteful. And this can point to the idea of perverse incentives rather than bad people. A much stronger argument.

Gavin Runeblade 12 May 2026 16:56 UTC
4 points
0
in reply to: Mo Putera’s comment on: Mo Putera’s Shortform
That really doesn’t sound like the inability to visualize. Whatever it is it seems to be something else
Here’s an example. My dad was talking to my youngest brother who was saying he didn’t like the Lord of the Rings movies because the actors didn’t look like the way he imagined them. My dad got confused and asked what my brother meant. My brother said, ” you know, when you’re reading and in your mind you see a stage or theatre and the people are acting out the scenes you’re reading? The people in the movie look different than the people in my head. ” My dad turned to me in shock and said, “I’ve never heard of this, do you have a little theatre in your head? ” And I said that no I don’t. My brother and dad kept talking where my dad was asking for examples and details of what it was like to imagine things visually. I lost interest because I had long ago realized my friends thought the line about “try not to visualize a pink elephant” made sense and I had long since stopped asking teachers why they thought that lesson worked.
I dream, but there’s nothing to see. It’s just narrative and touch. The me in the dream isn’t blind, and can interact with things, but the me that is conscious of the dream can’t see. I know the plot, I know what the dream me is reaching for. That’s just awareness not vision.
If you ask me what color pink is, I know what to say, but I can’t see it unless there’s some around us I can point to.
This is lack of visualizing. Whatever Ross is experiencing goes far beyond this.

Gavin Runeblade 11 May 2026 19:41 UTC
8 points
0
in reply to: ChristianKl’s comment on: Who Got Breasts First and How We Got Them
Mongolians have almost the lowest tolerance for milk among human populations, and the absolute highest rate of dairy in the diet. They developed cooking techniques that overcome the intolerance to achieve this feat.
Another confounder in this area is much more modern: most claimed incidences of “lactose intolerance” are actually intolerance of protein changes from the machine homogenization process and flash pasteurizing process
The key research was on people who can drink goat milk which has higher lactose levels than cow, and people who can drink raw milk.
Not saying the consumption of milk in Europe is a coincidence, just saying seek larger amounts of evidence than you might initially expect to need. This area has a lot of noise hiding the signal.

Gavin Runeblade 8 May 2026 0:47 UTC
1 point
0
on: AI #167: The Prior Restraint Era Begins
I need some context on the H200 order from China, how much is that vs the gap with America? And is NVIDIA actually going to deliver it? Have they already said yes?

Gavin Runeblade 7 May 2026 21:59 UTC
2 points
0
in reply to: Against Moloch’s comment on: Monday AI Radar #24
There’s very hard to calculate effects in the area. What has me surprised, is that robot arms have been around multiple decades and taken tens of millions of factory jobs. There have been robots that can do similarly specialized tasks, like I saw one doing drywalling back before COVID, but they’ve taken at most hundreds of jobs in almost a decade. Possibly only dozens.
They seem to have the right servos for strength and motion, from the videos and interviews I have seen. That suggests it is a software issue. But every model needs different software. That seems like a small.barrier to current AI coding tools, and a negligible speed bump for future coding tools.
But I don’t have enough info to sort hype from reality. Especially with the Chinese robots.
I only know one thing: these numbers are far too low. They assume factory-style build out. Ukraine is out producing the entirety of Europe with robots made in basements by tons of small companies.
This is going to spread, it is basically how the drug cartels make submarines and fast smuggling ships. So the model is proven in different industries. I expect it will be all over South America, the US, and Australia. And once India gets in on it the numbers will be enormous. I don’t see limits except for raw materials and energy.

Gavin Runeblade 6 May 2026 19:10 UTC
4 points
0
on: Monday AI Radar #24
They conclude we could plausibly produce 1.5 million to 3 million humanoid robots per year by 2030 (perhaps up to 10 million if everything goes just right). That’s an impressive number, but not remotely enough to put more than a small dent in overall employment.
I think that the cumulative number is enough. From the linked document:
Naively extrapolating these production trends to 2031 suggests a world with ~75 million humanoids, ~24 million quadrupeds, ~312 million drones, and ~593 million wheeled robots. The projections for quadrupeds, drones, and wheeled robots seem plausible if current pilot programs in surveillance, delivery, and logistics prove successful. The humanoid extrapolation is much more speculative, since the current growth regime is probably anomalous.
The labor pool in the usa is 170 million, roughly. Per FRED.
If half of those humanoids are in the usa they would be 18.7% of the combined pool. Maybe less than half will be, but we are the third largest country, not a lot of places can absorb that many millions of workers. China and India can absorb the units, but I wonder if they can afford to buy the robots before usa does.
I also note the calculations assumed no factories would retool to make different types of robots. I assume upgrading from robot arms to humanoids, has already got quite a bit of the supply chain baked in and is easier than a new fan from scratch. But this may be a bad assumption.

Gavin Runeblade 1 May 2026 19:19 UTC
1 point
0
on: Housing Roundup #14: You Can’t Build That
Interesting comment about Hawaii. I wonder if Yglasias meant that we (Aloha) have less obstructionist or just less Republicans. The difference matters for guiding other states.

Gavin Runeblade 13 Apr 2026 20:09 UTC
1 point
0
in reply to: Tim H’s comment on: Claude Mythos #2: Cybersecurity and Project Glasswing
On top left by the X in a black box it says May 2026. It is a little confusing with the line graph only going to April, but the overall snapshot is May.