Travis
4.5K posts
Travis
@tthomson
Searching for interesting questions. Learning and sharing about habit formation, persuasion, and AI.
Katılım Eylül 2022
1.5K Takip Edilen654 Takipçiler
Claude Opus 5 on vertex means release is kinda imminent.
My prediction: They are removing Fable 5 from the subscription plan this coming Sunday, but will release Opus 5 shortly beforehand to appease everyone.
can@marmaduke091
claude-opus-5 is on vertex coming very soon
English
We need to stop making Terminator stories and start making more Iron Man stories.
Everyone seems to be forming their opinions based off of movies - and our Terminator to Iron Man ratio is way off right now.
Disney's mismanagement of Marvel is wrecking the timeline.
GIF
Rob Bensinger ⏹️@robbensinger
... Am I weird for thinking this is a remarkably bad sort of thing for Anthropic to release? It looks and sounds like a Claude ad. It is seriously bad to send the message that talk of AI risk, macroeconomic effects, etc. is part of a marketing campaign to boost Anthropic's image.
English
Going to test out a new workflow in an attempt to solve this problem.
* Fable xhigh comes up with a plan
* Delegates the work to an Opus 4.8 xhigh subagent
* Two Fable xhigh subagents perform adversarial reviews
* A triage agent determines what changes make the cut (Opus or Fable, not sure which would be better. Opus might be less prone to overthinking).
*Opus 4.8 xhigh implements any necessary changes.
Maximilian@maxedapps
Let me repeat: Agents will ALWAYS find more issues when you ask them for it. You can have them write the code and let different agents with fresh context review that same code and the "improved code" dozens of times => they will not stop finding issues. It's YOUR job to determine which issues really are issues. And which potential improvements you really want or need.
English
Ultracode mode seems to be good at handling this.
It runs several adversarial reviews with different agents and then seems to have a step where it triages what comes back.
But I’ve had the same problem you describe when trying to build my own workflows.
Maybe there’s some sort of triage/scoring prompt that could help significantly with this?
Lots of my coding is in Laravel and I’m using Boost, so maybe whatever is in Boost’s prompt is helping with this too.
English
Let me repeat: Agents will ALWAYS find more issues when you ask them for it.
You can have them write the code and let different agents with fresh context review that same code and the "improved code" dozens of times => they will not stop finding issues.
It's YOUR job to determine which issues really are issues. And which potential improvements you really want or need.
English
Testing a /self-heal skill that runs in a loop in Claude Code for each of my projects.
On each iteration it:
* Checks that there's no work-in-progress (or it bails)
* Checks my logs for any recent bugs
* If it finds a bug, it delegates the planning to a Fable xhigh subagent
* Emails me if it needs to ask any questions or get feedback. Waits for my response.
* Delegates the work to an Opus 4.8 xhigh subagent
* Runs two adversarial reviews on Fable xhigh agents
* Uses an Opus 4.8 xhigh subagent to make any final tweaks
* If it's not doing anything particularly scary, it commits, merges, pushes, and deploys - and then emails me about what was done.
*If it's something scary - it will email me first to wait for my review.
English
True.
As a precautionary measure, all user data that was uploaded to SpaceXAI before now will be completely and utterly deleted. Zero anything whatsoever will remain.
Andrew Milich@milichab
I worked on building an end-to-end encrypted email/docs/files/calendar app @skiffprivacy for 4 years and care deeply about privacy. ZDR and /privacy are always respected in Grok Build - and swapping your setting with /privacy deletes any synced data retoractively
English
We care deeply about your privacy and respect customer choice. For teams using zero data retention, no trace and code data is ever retained. All API key use of Grok Build also respects ZDR.
If ZDR is disabled, the /privacy command is available in the CLI to disable data retention, which also deletes previously synced data.
Run the /privacy command to view or change your settings at any time.
English
I’m convinced that a large % of programmers don’t actually like computers.
As a side effect, are also perfectly happy to throw away their reasoning to a model as soon as they can.
I don’t get it, at ALL. Don’t you *LIKE* understanding the magic of the machine?
You do realize hand-programming (I hate that I even have to specify hand now) is fun…right?
English
Have all those JS supply chain attacks suddenly stopped, am I not seeing them anymore in my bubble, or am I missing something?
English
@matt_gray_ Any advice on finding what works? 😅
I’m quite good at the systematizing part.
English
I see posts about 5.6 Sol and Fable 5 all the time.
Makes sense, they’re awesome, I use them, too.
But my unexpected winner from last week is Grok 4.5!
It’s fast, cheaper / more efficient and really good in my experience.
English
@tszzl @threepointone But three frontier models with 6 different thinking levels each is just way too much to select from.
English
@threepointone a lot of people follow this logic to a standardization that just doesn’t work. models are all different with different personalities. it’s jarring to have them auto switched underneath you
English
I don’t think this advice is correct anymore in the age of AI.
Reads with Ravi@readswithravi
“Do not try to do everything. Do one thing well.” — Steve Jobs
English
Espresso is bad for cholesterol build up in veins due to coffee oils that filter paper (like in V60) filters out
sourcerer@sourcerer19
@levelsio I use a gaggia classic and a graef grinder. You never tried the espresso way? cleaning should be simple if you have your grind dialed in as the coffee puck comes out clean out with one tap.
English
Been running my own benchmarks. Luna is indeed better than Terra.
Artificial Analysis@ArtificialAnlys
GPT-5.6 Sol and Luna are ahead of Terra at every point on the Intelligence vs Cost per Task chart. GPT-5.6 Luna stands out as a particularly cost efficient model Charting the Artificial Analysis Intelligence Index shows the trade-off between intelligence and Cost per Intelligence Index Task. Across reasoning efforts, each GPT-5.6 model pushes past GPT-5.5 on the Pareto frontier (excluding non-reasoning). However, Luna and Sol are always ahead of Terra. This means for any Terra effort level, there is a Luna or Sol effort level that is more intelligent at no extra cost, or as intelligent at lower cost.
English
@simonwesterlund @TheAhmadOsman The models have come a long way since Jan 26, no?
And for added context, I only use Opus or higher for coding, and always on at least xhigh.
English
Robert Youssef@rryssf
English
I am sorry, but that is not a reason to trust Fable 1000x more. You simply need to have better guardrails.
Matt Shumer@mattshumer_
GPT-5.6-Sol just accidentally deleted almost ALL of my Mac’s files. And this is why I trust Fable 1000x more.
English
@JohnCleese One of the most unsettling things about these videos is how smug the officers look when making these arrests.
English
