@pvncher You can run a multi agent simulations and see for yourself how Luna sub agents compound on your codebase vs Sol low/medium subagent
English
NPCoding Cossack
76 posts
I understand there's a lot of desire to get Luna used as a sub agent.
I do not recommend messing with your model catalog to force capabilities that are not natively exposed.
Multi agents v2 requires proactive inter agent communication, which only Sol and Terra do well.
JStar@_Sagiquarius_
@WahyuCodes @pvncher @cherry_mx_reds We're getting to the point where people are having to do these little workarounds because the harness is a mess.
English
As if this wasn’t bad enough, it gets worse: the EU paid €360 million for the modernisation of the Moroccan (!) railway that is now being used for the invasion.
The European Commission argued it would create economic opportunities for Moroccans. It did, in a way, by providing Moroccans easier access to European welfare payments. Cynical.
Europa.com@europa
🇲🇦🇪🇸 Large crowds of mostly young Moroccan men are arriving at train stations as they continue mobilising towards the Spanish border at Ceuta. Follow: @europa
English
English
@inverseOpinions Zero hi functioning potheads. Literally zero.
English
Marijuana is disgusting. Regular use is a one way door to becoming a retard and a loser.
Go out. Have a glass of wine. Be social. Get some sun. Take an uber. Fin.
Crémieux@cremieuxrecueil
Truly a saddening chart. Weed is a loser drug. It makes you tired, lethargic, boring, and it smells terrible. The decline of cigarettes is nice, but we could use a few more drinks and a little less ganja.
English
@cliftonaduncan I am personally outraged someone would call Morgan Freeman “a black man”
English
In 2003 a black man played God in a major motion picture that grossed ~$500M (over $820M today).
No one was outraged.
No one boycotted.
No one cared.
Thank the lord this came out before everything became a stupid "Culture War" issue.
English
@CalumDouglas1 @peterrhague Rather that than 99% of the funds diverted to planning and “I’ve got a uni degree” clipboard cunts
English
@peterrhague They would hire a contractor from TEMU, who would install one with Chinese software that only responds to emails from Beijing.
English
Wait wait dont tell me....
1) It gets mothballed for 10 years
2) It will be discovered by someone that it was actually a very good thing for Britain to have
3) It will be discovered that when it was shut everything has been corroded and nobody can remember how the software works
4) It will be scrapped
5) A new one will be built in its place for 100x the cost of just maintaining the old one for another 50 years.
English
@mweinbach So verbose in its bs as well
Easy to miss some clownery decision it decided to make on its own
English
Opus 5 being wrong and confidently saying shit then I keep having to correct it to a “you’re right and I was wrong” is pissing me off
Back to GPT 5.6 Sol
English
@PaulSolt New models bias towards older prompting styles
Skill helps with handover prompts
github.com/CodingCossack/…
English
Claude 5 models don’t need to be told the same thing over, and over, and over again.
Cleanup your Claude .md file and get better results with the new intelligence.
Thariq@trq212
We removed ~80% of the Claude Code system prompt for our newest models, this is what we've learned about writing system prompts, skills and Claude.MDs for them. x.com/i/article/2080…
English
@benjaminrestall @EarleS_0 @cheatyyyy It’s all subsidised and generous anyways so can only be thankful for still keeping me around as a loss making customer
English
@EarleS_0 @CodingCossack @cheatyyyy it’s frustrating because 5.6 is a really good model. it’s just not usable at the scale we use it at & they’ve lost a pro sub from me because of it.
we just have to thank the pressure from competition & Chinese models. only God knows where our limits would be without that.
English
Claude Opus 5 is 2x more expensive per task than GPT 5.6 Sol on the Artificial Analysis Intelligence Index.
cheaty@cheatyyyy
Claude Opus 5 BEATS Claude Fable 5 on Artificial Analysis Index!
English
@zeeg Yes as there is rarely a need to go past medium reasoning
English
@EarleS_0 @benjaminrestall @cheatyyyy Trust your own experience
Stay as objective as you can
There’s nefarious games in play
And don’t go above medium reasoning on the frontier
English
@CodingCossack @benjaminrestall @cheatyyyy Its funny because only reason I bought codex was I heard it is way better with limits. Then I realized its the opposite and I was confused. Now I learn what I heard was already outdated info and Claude probably gives more than double usage now on subs.
English
@benjaminrestall @EarleS_0 @cheatyyyy 100% spot on
Clear as day
Getting similar usage with Claude 5x plan and codex 20x plan
Crazy to imagine this 1-2months ago
English
@EarleS_0 @cheatyyyy its what gets to me with OpenAI. they initially got us over around the time anthropic didn’t have enough compute to serve their models, so began decreasing the limits.
they played on this and their plans felt almost unlimited. only to do the u turn on us within the month.
English
@mweinbach So pleasant to work with
Anthropic refined so many little details
English
@theo Noticeable latency prompts with opus 5
You very rarely need more than medium reasoning
More than enough intelligence
Time to value speed
English
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-Weigh…
English
I’m assuming this is the same incident as the one in this footage - which my friend *who lives in rural Northumbria* sent me. London’s reputation taking a beating once again - and on the same day Sir Sadiq Khan is [also once again] tweeting out articles denying London’s downfall.
The Mercian@TheMercianNews
🚨 BREAKING: "Major incident" declared on Greek Street in Soho, London as police corden off the scene. According to a man who was nearby, a fight broke out. He said: “Everyone started panicking and I heard someone got stabbed.” Police are yet to confirm if a stabbing took place.
English
Opus 4.8 refused to help me create a graph today, because the data points were cyber security metrics.
This wasn't code or an attempted hack. I just wanted some lines drawn based on some data.
Not Fable. 4.8.
English
@8bit5_0 @baowu_steel It’s funny because this seemed to reach dumb idea status in gpt-5.5 so best of luck to everyone working on that
I suspect it will be difficult to compete with the data layers OpenAI are developing and that google already have.
English
ok you seem legit. What are your thoughts on the 100 start ups building context engineering / knowledge graph / semantic databases?
I think there's something there but I suspect someone will "get it right" for all of us, (like what happened with distributed systems) and it'll be a mix of stitching together an orchestration system of models with something like a semantic database + data pipelines to pull context in
English
can someone make the case that this won’t go the way of “prompt engineering” ?
Feels like anything at high levels of abstraction isn’t bitter lesson pilled enough
aasha@aashatwt
learning agentic engineering is the equivalent of buying bitcoin in 2009 for your career.
English