“ This is not the smoothness problem solution, but rather the Burgers vortex, an exact solution of the Navier–Stokes equations published by J. M. Burgers in 1948, chosen because it has the same anatomy: fluid drawn inward in a plane, stretched along the axis and thrown out of both ends, spinning fastest in a core.”
To my mind, this blog post had the same stilted, mechanical feel that those all talk YouTube commercials have, even when they're in an Australian accent. Nothing very bad or very good, just mid phrasing and vocabulary.
Funny to see that they did not include Fable 5 in their GeneBench and LifeSciBench comparisons because "it does not answer advanced biology questions and refuses the majority of questions in this eval".
It's so absurdly sensitive. It bailed out earlier today working on a TypeScript client for a sensor network API which happens to include some temperature and pH sensors for tanks, which yes, are used for biology experiments. But wow, we're degrees of separation from the actual biology work.
It's making it very hard to justify even trying to use Fable. When it works, awesome; it's legitimately good. But I can't trust it to do a task without deferring to Opus and that's really annoying at times. I want to know what I'm getting up front, not after the fact.
It refused to give me plant care instructions for an ornamental sold at my local Home Depot because it decided it was highly invasive and dangerous to grow in my region.
In their defence, google searching anything about plants these days leads to awful results. It’s saturated with slop. A response from an LLM might be slightly more reasoned and targeted. It’s hard to tell. This is a category of knowledge that’s being destroyed by people gaming google and dumping huge amounts of bad LLM and image generation onto the web.
This is also true for pretty much any other category of search that you do as a layman. Specific, targeted queries for official documentation or research are fine, but if you look up basic information about cars, health, or basic computer troubleshooting a massive portion of the results are AI-generated.
If I'm going to be getting AI-generated results, I'd rather read the output of a model that has all the context of my specific situation than slop generated in bulk with a cheap model from six months ago.
I asked whether an outdoors mosquito trap product (via a screenshot) would negatively impact other insect species in my garden and it refused. Though quick internet search did reveal that it would harm and trap many other species of harmless insects.
I asked him about sharks to be able to answer my kids question and it got triggered somehow. Then again when I asked it if my code had bugs or vulnerabilities before I commit.
At some point just kill the thing, it's not able to work properly as it is.
I'm writing a programming language with a "capability security model". That's enough to trigger Fable, it won't work on the language. It's hilarious. The mere presence of the word "security" seems to be enough to trip it up.
Anthropic refuses to allow Fable to code review my interpreter's memory safety. It was funny at first, then it became disappointing, then insulting, and finally utterly infuriating because I remembered the fact I'm actually paying for this nonsense.
Cancelled my subscription today. Hope OpenAI isn't patronizing like Anthropic. I don't want to hear about their "safety" bullshit ever again.
Yes it has completely turned me around - was all in on Anthropic but now it just looks too risky. Better off leaning into open models. Even if I found a way to work with the restrictions as they are, who is to say they won't suddenly change tomorrow. It's not worth it.
I mean it's a fucking joke, I kept getting refusals on a code base I wasn't familiar with and it was literally just because there are some vars named DNA. Just absolutely stupid.
Anthropic just refuses to allow Fable to properly code review my projects. It's so obnoxious. If OpenAI's Fable equivalent is better at this, that'll get me to cancel my Anthropic subscription and switch.
Given that Fable is so gutted and Anthropic added the absurd data retention policy for it, I'm going to advocate that we prioritize support for as many other models as we can at work.
The other day I asked Fable about fasting for 16 hours, and it flagged my question.
Pathetic situation, this one, where we are supposedly building a superintelligence while at the same time thinking that fasting is a biological weapon.
My new favorite new passtime with frontier LLMs, keep telling them "I just ate a [non-food object]". Eventually it gets stuck in a loop of telling me to call 911, unlock my front door, and lay on the floor in case I lose consciousness.
Well it seems like they removed quite a few 3rd party benchmarks they used for GPT-5.5 release where Opus 4.7 was better and added many new benchmarks created by them where conviniently GPT leads.
I recently asked Claude to help me choose a single MOSFET (transistor) for a specific use case in a mundane circuit. The safety triggered and it ended the conversation and refused to continue. Gemini has also done the same thing to me. Looks like the big players got very spooked by the temporary Trump admin ban on Mythos and they all locked down way too hard.
It triggered for me on a completely pedestrian game design prompt a couple of days ago. I’ve sent feedback and continued with Opus, but that was really unexpected
The microbubbles in scuba diving that cause the bends are the ones trapped in joint space fluid. That fluid doesn't circulate at a useful rate, so unfortunately you can't really "burst them somewhere else" =(
While diffusion from joint space is perfusion limited, some work suggests ultrasound might increase that perfusion? By vasodilation and microvascular recruitment, from heating and shear stress on endothelial cells. With bubble vibration and cavitation as one source of stress.
Also, though perhaps not rate limiting, ultrasound might be able to mess with the N2 bubble boundary diffusion rate.
Caveats: Very not my field; no clinical practice; mostly animal studies; mostly musculature and not joint; and under-validated AI.
Meta: But I so very much enjoy surfing literature with AI. Even with AI's rich collection of interesting failure modes. They serve as fun added encouragement to keep you on your toes, and keep clear on the gradients of your confidence.
This is a great puzzle game. I think it actually teaches the concept of "piece coordination" in chess very well.
One suggestion is to have all enemy pieces move simultaneously. I expected that the losing condition is that I am threatened and there are no unthreatened squares to move (checkmate).
However, since the opponents move one-at-a-time, I found that even if I moved to a safe square, sometimes I could be both threatened and captured in the same move! Which is somewhat different from normal chess, since now you have to consider the possible orders in which the opponents move. So even moving to unthreatened squares could be a game over.
You can click your own piece and see the move order to see if it's a possibility you'll be moving into a discovered attack. You can sort of predict that pieces will move if they can go from a non-threatening state to a threatening state.
In real-life, before the election there is a margin of error on the support of a bloc.
If you interpret a "tie" in this game as "either party could win within the margin of error", then it becomes a lot closer to solving the problem that gerrymandering algorithms try to solve in real life!
Have you played the game? The board today has three parties and the winning solution is to make sure your party wins one district and the other parties tie in the other four.
Under any real world system, you will lose this election if you rig it this way. You just can't predict who will win it instead.
Edit: Per a comment below, this does not seem like a regular feature of the game, just an oddity of today. It would still be worth figuring out a way to eliminate the tie issue or at least ensure it's less of a factor in future games, but the game is much more fun on average overall than today's game suggests
Yeah I don't know if that was just the puzzle today, since this is the first time I've heard of or played this game, but that feature seemed like a disappointing execution of an otherwise genuinely unique idea.
Winning a single district for your extremely minority party while locking the other two parties out of winning anything isn't even remotely analogous to how real world gerrymandering works, at least in the US where the term is typically used. It also feels like cheating, since it relies primarily on exploiting a flaw that exists exclusively in the game but not in real life. I'm all for simplification of real world factors in games, but not to the point where the entire path to victory relies on that simplification.
A more accurate and more interesting variation would be to just have two parties with puzzles that rely on crafting districts where the party with fewer voters wins the majority of seats, with the challenge coming from voter distribution patterns that make it hard to create winning districts while following the game's rules. The addition of a third party and ties that result in nobody winning seem both unnecessary and worse.
The goal of this puzzle appears to be to get more people to talk about the issue and push for change. In these things, you can have either popularity or nuance but not both. The average American can't even read, let alone understand the nuance or complexity in how gerrymandering "actually" works.
The game stores and allows you to see the RNG seed that controls the run's events and layout. The developers want players to be able to share seeds that produce interesting runs.
That requirement is what made this problem difficult for the devs to solve.
This shouldn't actually be difficult to solve though.
The issue is that knowing the offset of seeds helps predict outputs.
Instead of calling RNG(seed+hash(string)) 10x, make one RNG(seed) and call that 10 times to get random seeds for your 10 rngs. Now you have perfect determinism and no correlation.
My first solution was RNG(hash(seed.toString() + string)), which would get rid of the correlation while still being deterministic based on the seed.
It's also more robust than calling RNG 10 times since if you use the same algorithm to seed as for the RNG proper then you will get the same sequences in each instance, just offset.
That's assuming the game initialization order is deterministic. Using the hash of the combined state of seed and string avoids that assumption without giving up determinism.
Point being, the current problematic state of the game is trivially fixable in multiple ways that require half a second's thought (once being aware of the problem).
The only reason I think biotech companies are not yet raising hell (and invoking the False Claims Act) is that Thermo Fisher's antibodies are already known to be notoriously bad, and everyone serious seems to have to validate everything themselves.
I'd treat this about the same as datasheets for mechanical or electrical parts.
When I buy an electronic component as a regular consumer I expect the datasheet "typical" values to be accurate 90% of the time. I can imagine larger industrial customers would really raise a stink if it's worse than that. However, any critical components in my circuit must be verified and "binned", and that's on me.
This is the thing. Yes, the marketing material is bad. But, no one in lab trusts an antibody just because of where you bought it. A new antibody always gets tested and validated before use.
That is to say, this looks bad for Thermo Fisher. But, that’s as far as the damage should go.
Why would you even generate fake pictures of this type? Don't you already have real ones? I mean, it's actually more work, unless you don't have the real ones.
I’m not going to defend Fisher here. It was a stupid thing for someone to do.
But unless you’re in the field, you won’t realize exactly how big ThermoFisher actually is. They are the major supplier of everything for molecular biology work. From freezers (the Thermo part) to plates and pipettes (Fisher) to enzymes and antibodies. In many ways they are like Amazon. They sell everything. Some of it from outside companies, but a good deal of sales are from in-house brands. They could use their position as a reseller to know which products sell the best and with the highest margins.
In a company of this size, it’s easy to have one group feel pressure and cheat on running the gels to confirm results. Particularly when the real results are ambiguous or dodgy. It’s not a good look, but I doubt it will put a dent in people from buying things (non-antibodies) from them.
reply