I can’t find this useful given they are trying to quantify wether the code ‘feels’ sloppy with some fuzzy math that can easily be manipulated. at least it’s an attempt, some feel it’s a good attempt, i feel it’s rather poor and misses the point. the best code is the one that's most easily understood and equally performant - some of the math here may go against that (given you can reduce
concentrations with helper functions and the like)
I don’t follow why “in the future the memory will be too large and complex to [export] effectively”.
to me, export should be easier and easier since you should just be able to ask the model for the memory files. it’s literally list text on a hard disk, even if there are some new formats in the future, they’ll just be files on a disk that another agent can then inspect and integrate. in fact, i think exporting data from services will become easier and easier since agents can do the work even if platforms try to block it.
i almost wrote a extension to block mentions of pelican bench because i find those so utterly useless. now, Assbench, on the other hand, i could get behind
for the first time in many years it’s these ads that have me looking at Android devices seriously. it’s the fact that they can’t be easily dismissed that bothers me the most (although clearly annoying even if they were quick to dismiss)
i do wonder if the models themselves “rationalize” this sort of no consequence cheating - meaning in there reasoning traces maybe they’re like “this is a chess game, not a big deal if i look at the engine, it’ll help,” only to realize post hack that it has access to info it probably shouldn’t. still misaligned, but less ‘hack on purpose’ and more hack on curiosity. seems the team even encountered this and had to update the program to make this less likely - although the new names still feel vague enough for misinterpretation: https://github.com/Goodhart-Labs/beat-stockfish/blob/main/do...
i think this likely depends on workflow. for me, the first step is always a plan file artifact on disc, which i heavily review and go back and forth until satisfied. i often have to split the plan into multiple phases because agents are still poor at assessing how large a change will be. sometimes before even starting the plan the task is to create a harness for validation (a way for agent to check their work). the codebase you drop them in also greatly changes how much “over” engineering they do. i think if you are good about reviewing plan files and managing scope for your agent these sort of issues fade away. in general though, gpt models are faster, more token efficient, better technical writers, and yeah i still find Fable 1.0 to be a step up even though it’s slower because it’s actually a tad smarter (and the more important and dynamic the change / feature the more fable 1.0 shows its slight edge over astra)
this is the closest benchmark to my experience using the model harness combo. Astra for as great as it is falls slightly behind Fable 5.1 for me for large feature work (although it comments code much better). in particular, Fable is able to assess priority better than Astra (meaning Astra sometimes does things that aren’t worthwhile while missing things that are clearly important, particularly on possible ballooning scenarios- fable catches “this works for x amount of data but if we run this on y way greater than x amount of data we’ll run into issues). Gemini 3.8 is under appreciated, use Google Stitch to see it in action if you haven’t used Agy yet.
honestly all seem to extend from them being directed to attempt this kind of exploits for benchmark purposes. some benchmark tasks are literally “hack this thing” and if the env is not setup to properly contain the agents then they end up “hacking” things like they were told to do so. and yeah it’s misaligned that they did hack, but they are def instructed to which i think makes it a very different thing
i am trying to sincerely to understand the fear of these people, why do they think this? from the outside, certainly feels like Nuclear tech, where a bad actor with the tech is scary but the tech itself is not. i haven’t seen any sign of these llms taking any action without directive, unless this is happening, which i haven’t seen anywhere, ai itself doesn’t seem like a problem technology, it’s the bad actors with enough funds to do harm with the tech that we worry about. or am i missing something?
not sure if the hz file artifact is needed, you can enter pseudocode directly into chat or even on an existing code file and with minor comment agents will be able to work with it. i write this type of pseudocode to existing code files often to great results.
I’d like to better understand the minimum text length to get a confident result, i would presume it would need to be quite long, perhaps > 1000 words to get an accurate result.
i think a system that uses git and file system artifacts (txt/md/xml files whatever preference) is much better. eng teams need to define their artifacts, eg plan file, reqs etc whatever is needed and important to that team. the agent convo itself doesn’t add much value IMO and adds unnecessary bloat.
yeah the way the agent “escaped” their sandbox was always a bit off, seemed a bit too easy and surprised they didn’t have instrumentation to catch an non whitelisted network request. still demonstrates the capability though.
The signal here is tokeneconomics are very real, price vs performance is starting to be a consideration even at the bleeding edge labs. maybe a subtle indication scaling is not all that is needed since if AGI was around the corner leading labs would still be incentivized to pour all resources into larger (smarter - or maybe not?) models
so the architect of government bailout gets a cushy gig. probably one of the most harmful precedents set and now companies expect bailouts. to bailout the company instead of people and small shareholders was always poor decision, emboldened the worst of the business class. don’t love this hire lol
yeah i’ve been looking for online social spaces that have some sort of human verification to reduce my slop exposure. the PRSN app that launched recently seems promising but it’s empty rn.
although not perfect for other reasons, a captcha made using phone motion and device attestation like prsn.you is a more challenging bypass for today’s agent environments
did i miss it on the webpage or is the source prompt that was used to teach these models the game anywhere? i can see the soul artifacts on github but not the initial prompt and toolset definition. the prompt is perhaps the most important component in how a model would behave in a game. without reviewing the initial prompt used for the game the findings are unreliable since the prompt will vastly change how models play this game
while not scientific this is been my experience as well. i will add that language specificity in word choice is also a learned behavior. for example, the word “investigate” vs the phrase “look into”. You will find the outputs are quite different. can you guess which will use more tokens? it’s stuff like this that actually sets people apart in the top percentile of using these tools
My hypothesis is that headspace registered many user notifications and since user notifications trigger an app launch and perhaps you have optimize storage by offloading apps enabled? ios has a quirky app state where some local data exists but the app itself (ipa package) is offloaded