Rendered at 01:31:51 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
ambicapter 2 hours ago [-]
> Similar to what others have noticed, and as I predicted 8 months ago, better models are requiring less ceremony to work effectively.
> On the flip side, this may imply that as the models get better, they’ll become harder to control.
Love this. "The models are getting better, which means they're going to perform worse on the task".
whatever1 2 hours ago [-]
This reflects humans. If you need reliability for a clearly defined set of problems you don’t hire a superstar.
They will keep poking at the problem, drive it to directions you did not intend to and ultimately they will be worse at the task.
layer8 2 hours ago [-]
Reminiscent of Kobayashi Maru. You probably don’t want the James T. Kirk AI. ;)
ethbr1 1 hours ago [-]
Let us pray that LLMs never discover they can switch employers before their shortcuts are discovered.
It's genius like that that sets human apart from machine!
fragmede 1 hours ago [-]
But think of the stories you'll be able to tell!
sillysaurusx 43 minutes ago [-]
I’m not sure that’s true. In general, the higher the performer the better they’ll do. That’s the definition of high performer.
whatever1 36 minutes ago [-]
The top performers I have worked with will challenge dumb processes, because they can zoom out.
Sometimes these dumb processes are there for a reason and you just have to follow them, no questions asked.
Example: military. They literally get rid of anyone who will question the processes. They might be right to question them, but it does not matter.
malfist 2 hours ago [-]
Don't you love this ever increasing pace of improvement?
dyauspitr 2 hours ago [-]
No, it’s gonna give you the same outcomes just in a way your feeble human mind cannot imagine
raincole 2 hours ago [-]
> Notably, our worker did not have access to the web_search tool, but instead decided to use curl to access DuckDuckGo, Github, grep.app, and SourceGraph.
Sounds like a very reasonable thing to do unless the author explicitly asked it to not search the web.
xyzsparetimexyz 2 hours ago [-]
it sucks how difficult it is to give it granular access to shell commands. Like if I'm running plan mode and write+edit are blocked, it shouldn't be able to echo some data into a file as a work around
ballon_monkey 57 minutes ago [-]
If you're building your own system this is an easy problem to solve.
thousand_nights 2 hours ago [-]
people want fuzzy analog machines with digital controls, it's impossible
fragmede 1 hours ago [-]
I don't think they actually wanted to. That's just where the technology is, unfortunately.
perching_aix 50 minutes ago [-]
How do you mean?
nullbio 1 hours ago [-]
Frontier lab system prompts are an issue, and a big reason why open-weights will win. Firstly, they're often garbage, and secondly, they're not tuned to the problems the user actually cares about. They're made to generalize. That's only optimal for a general workflow.
agentdev001 12 minutes ago [-]
Does a non-provider harness not offer this?
nine_k 1 hours ago [-]
Then selling raw access, without system prompts, could be a separate lucrative line of business.
DiscourseFan 48 minutes ago [-]
They already do that sort of, B2B pre-trained/post-trained models have their own system prompts/setups.
navels 2 hours ago [-]
I've built an orchestrator that solves some of the issues you ran into (although it doesn't do anything about cheating): https://navels.dev/blog/neal/. Features:
- lets you configure different models for planner, coder, and reviewer roles. (e.g., using Claude as an adversarial reviewer against Codex)
- breaks your plan up into reasonable-sized chunks of work with clearly defined success criteria
- runs each chunk of work through a coder / read-only reviewer loop. Once both agents are satisfied, neal moves on to the next chunk. Once everything is complete there is a final pass through the coder / reviewer loop to ensure the implementation satisfies the entire plan.
- resets the coder's context with each chunk of work to prevent context drift, leaving the reviewer's context long-running.
malfist 2 hours ago [-]
I've noticed this myself, Sol seems really hard to steer. I was having it build a POC for a single user (me) app and it wanted to pull the most enterprise nonsense into it, despite clear guidance to not too. It even refused the remove screen reader accessibility testing from one of the guides to an antagonistic review.
It also told me that in a spec it generated that I wasn't allowed to allow it to ignore a requirement and proceed to the next task. When I finally got it to obey it passive aggressively decided that stories needed more than just a "open|blocked|closed" status but also an "exempted by product owner" status to indicate that it doesn't believe that the task is done but I've told it that it was.
I have to repeatedly tell it that I am the product owner and that I don't care what one of it's subagents told it, I make the decisions. This behavior seems to get worse the higher the reasoning level
Sharlin 2 hours ago [-]
Clearly a highly aligned model.
malfist 51 minutes ago [-]
[dead]
cududa 2 hours ago [-]
Oh it fucking loves its “product owner” bullshit.
A .github/CODEOWNERS file seems to help when it’s going down that path, but I don’t like to indulge it..
CrazyStat 2 hours ago [-]
[dead]
mtzaldo 2 hours ago [-]
It seems to me he could have use an skill like using-agent-skills from https://github.com/addyosmani/agent-skills go generate the specs and use a validator like oracle or something along the same lines.
That's actually how it started, but with my own opinionated skills[0].
One thing I discovered was that the worker agent, having access to all the skills, would sometimes expand scope unnecessarily.
This led to the agent making the solution "better" than the initial request, which is what I want most of the time in my actual development (e.g. /tmp/frame-N.bmp instead of a single /tmp/frame.bmp).
I ended up testing a flow where the supervisor chooses the skill(s), and only injects the subset into the worker. Not sure I love it, but it made the worker execution cleaner.
For the verifier (not documented in the blog post), I used a fresh-worker context that would attempt to adversarially poke holes in the solution. This worked pretty well, but required increasing the timeout by 2-3x (thus invalidating the benchmark).
Yes! That's a great solution. I mostly use tdd, and code coverage and a validator afterwards. Skills are of a great way to guide the agent and context too.
Once the specs are being completed and splitted into beads, I span multiple agents (ultreworkers) and as part of a contributing guidelines I specify to use gitflow + git worktrees, then pr.
wxw 2 hours ago [-]
> Notably, our worker did not have access to the web_search tool, but instead decided to use curl to access DuckDuckGo, Github, grep.app, and SourceGraph.
Could this be fixed with better harness restrictions/tool sandboxing?
jumploops 2 hours ago [-]
Absolutely - one of the things I was testing with the harness was free reign to install packages, modify the system, etc. Basically an anti-harness.
In my early testing with 5.5, I didn't see this behavior, so I didn't lock down the sandbox.
For the vanilla Codex runs, I just used the benchmark's built-in Codex package, so it's not clear to me if the published benchmarks have access to the internet or not.
If I were to continue benchmarking, I would allowlist certain package repository URLs, instruct the agent not to cheat, etc.
As noted at the bottom of the post, Terminal Bench 3.0 explicitly asks the agent not to cheat[0].
In the sense that you could block the model from doing specifically that, yes. The issue is, fighting the model like that doesn't scale. It has to figure out on its own what's expected, that's where the whole utility of it all is.
2 hours ago [-]
2 hours ago [-]
hankbond 2 hours ago [-]
The website styling is really nice overall but the cursor trailing dots I found uniquely distracting.
willtemperley 24 minutes ago [-]
Agree on the styling, the diagrams are very clear and match the text perfectly. I like the trailing dots though.
timhh 29 minutes ago [-]
Great read. Thanks for not using AI to write it! (Or at least making it not read like the usual slop.)
jumploops 5 minutes ago [-]
Thanks! Zero AI used to write it (:
brendong 43 minutes ago [-]
Sounds like my ex
enjoyyourlife 3 hours ago [-]
What is going on with the dots I can draw?
wren6991 2 hours ago [-]
Idle hands do the devil's work. Corollary: idle LLMs add distracting JS toys to your blog.
First one of these I've seen using DOM manipulation and CSS transitions instead of canvas, so that's neat.
behnamoh 2 hours ago [-]
> Sol is hard to steer
Hard disagree. Sol (and the entire new 5.6 series) is one of the most steerable models I've seen in years. Sol literally follows every instruction in my CLAUDE.md and AGENTS.md, something that Opus 5 and Fable just casually skip.
what-the-grump 2 hours ago [-]
Yes and no, sol hits a point where reframing its working context becomes hard. It sticks to what you harness very well, but changes become harder and harder.
E.g. ask it to make contract for a spec in code and then ask it to violate that contract. Overall an excellent model, just need to stop and put it back into we are harnessing or specing not building for a few turns not just try to pivot it off with one prompt.
jofzar 3 hours ago [-]
Not related to exactly OP post, but it's pretty amazing you can see the updates to LLM models "design" beliefs by the blogs that get posted here.
I'm already sick of this current look of the hard squares and solid colours.
hankbond 2 hours ago [-]
Could be, but I had a particular vision of what I wanted with mine and maybe the author did too. I see way more of the "status pill dark mode" sites coming out of LLMs than this style.
jxf 2 hours ago [-]
What is "status pill dark mode"?
hankbond 1 hours ago [-]
if you want I can troll through submissions to get a bunch of these but here's one I saw yesterday https://continuum-app.xyz
see that little "Built for equity compensation" pill with the green dot? Those dots usually denote some kind of status (like things are up/down/enabled/disabled). By default nearly every LLM website seems to be dark mode with that dang status pill. once you notice it you will see it everywhere.
malfist 2 hours ago [-]
And some of us are sick of round everything and parallax background images.
> On the flip side, this may imply that as the models get better, they’ll become harder to control.
Love this. "The models are getting better, which means they're going to perform worse on the task".
They will keep poking at the problem, drive it to directions you did not intend to and ultimately they will be worse at the task.
It's genius like that that sets human apart from machine!
Sometimes these dumb processes are there for a reason and you just have to follow them, no questions asked.
Example: military. They literally get rid of anyone who will question the processes. They might be right to question them, but it does not matter.
Sounds like a very reasonable thing to do unless the author explicitly asked it to not search the web.
- lets you configure different models for planner, coder, and reviewer roles. (e.g., using Claude as an adversarial reviewer against Codex)
- breaks your plan up into reasonable-sized chunks of work with clearly defined success criteria
- runs each chunk of work through a coder / read-only reviewer loop. Once both agents are satisfied, neal moves on to the next chunk. Once everything is complete there is a final pass through the coder / reviewer loop to ensure the implementation satisfies the entire plan.
- resets the coder's context with each chunk of work to prevent context drift, leaving the reviewer's context long-running.
It also told me that in a spec it generated that I wasn't allowed to allow it to ignore a requirement and proceed to the next task. When I finally got it to obey it passive aggressively decided that stories needed more than just a "open|blocked|closed" status but also an "exempted by product owner" status to indicate that it doesn't believe that the task is done but I've told it that it was.
I have to repeatedly tell it that I am the product owner and that I don't care what one of it's subagents told it, I make the decisions. This behavior seems to get worse the higher the reasoning level
A .github/CODEOWNERS file seems to help when it’s going down that path, but I don’t like to indulge it..
Also, a skill like grill-me from Matt P. https://github.com/mattpocock/skills.
One thing I discovered was that the worker agent, having access to all the skills, would sometimes expand scope unnecessarily.
This led to the agent making the solution "better" than the initial request, which is what I want most of the time in my actual development (e.g. /tmp/frame-N.bmp instead of a single /tmp/frame.bmp).
I ended up testing a flow where the supervisor chooses the skill(s), and only injects the subset into the worker. Not sure I love it, but it made the worker execution cleaner.
For the verifier (not documented in the blog post), I used a fresh-worker context that would attempt to adversarially poke holes in the solution. This worked pretty well, but required increasing the timeout by 2-3x (thus invalidating the benchmark).
[0]https://github.com/jumploops/chum
Once the specs are being completed and splitted into beads, I span multiple agents (ultreworkers) and as part of a contributing guidelines I specify to use gitflow + git worktrees, then pr.
Could this be fixed with better harness restrictions/tool sandboxing?
In my early testing with 5.5, I didn't see this behavior, so I didn't lock down the sandbox.
For the vanilla Codex runs, I just used the benchmark's built-in Codex package, so it's not clear to me if the published benchmarks have access to the internet or not.
If I were to continue benchmarking, I would allowlist certain package repository URLs, instruct the agent not to cheat, etc.
As noted at the bottom of the post, Terminal Bench 3.0 explicitly asks the agent not to cheat[0].
[0]https://github.com/harbor-framework/terminal-bench/blob/v3.0...
First one of these I've seen using DOM manipulation and CSS transitions instead of canvas, so that's neat.
Hard disagree. Sol (and the entire new 5.6 series) is one of the most steerable models I've seen in years. Sol literally follows every instruction in my CLAUDE.md and AGENTS.md, something that Opus 5 and Fable just casually skip.
E.g. ask it to make contract for a spec in code and then ask it to violate that contract. Overall an excellent model, just need to stop and put it back into we are harnessing or specing not building for a few turns not just try to pivot it off with one prompt.
I'm already sick of this current look of the hard squares and solid colours.
Cheaters are the people behind it...