Tweet Disematkan
Luckey Faraday
647 posts
Luckey Faraday
@luckeyfaraday
🍀 Building Athena — context-aware AI coding tools that remember your decisions @luckeysystems YT https://t.co/iatVsRG954
Bergabung Şubat 2024
65 Mengikuti55 Pengikut
@trikcode Not everyone has money as their main objective
I’m building open source tools and don’t expect to make anything
Id rather give back to the community rather than extracting
English
What if we apply this same logic to small local models to try and achieve frontier-ish intelligence
OpenRouter@OpenRouter
Introducing the Fusion API, the smartest compound model in the market. Fusion achieves Fable-level intelligence at half the price. How it works 👇
English
@Atenov_D You should be pruging unused skills weekly or monthly
Why keep something you don’t use?
If you ever need it just redownload
English
A team of AI scientists just proved most "top" AI skills on GitHub are actively hurting you. 1 in 3 makes performance worse. Stars don't predict it.
> Here's what they did.
175,632 skills scanned. 19,214 pulled as candidates. 7,950 deep-indexed and run through real evals - not vibes, not stars, actual causal-lift testing.
The results:
1 in 3 skills make the task worse than running with no skill at all. 69 out of 199 measured skills had negative lift. Popular, highly-starred skills sit comfortably in the red zone next to garbage nobody uses.
Stars are not a quality signal. Twitter hype is not a quality signal. None of it correlates with whether a skill actually helps.
>> The deeper pattern:
The stronger the model, the less skills matter. If the model already crushes the task at SOTA level, a custom skill is more likely to break its style and context than add value.
Skills earn their keep on the hard stuff. Long pipelines, external tools, CRM, finance ops, complex GTM workflows. That's where the ROI is real.
>> What this means for you:
Stop installing skills based on GitHub stars or what's trending on X. You need actual evals, not internet reputation.
And even the good skills are competing with a $5K - 150K services market - freelancers, agencies, software.
In some zones AI recipes are already autonomous enough to replace that. In most cases you still need model + skills + real SaaS stack (CRM, email, billing) working together.
Full breakdown and the tested list: lib(.)cybos(.)ai/insights
Atenov int.@Atenov_D
FREE $100 in credits with access to Opus 4.8 and GPT 5.5. Sauna AI. No card. Gmail or Slack connect. Nobody is talking about this. Sauna AI gives you $100 in credits on signup. No card required. Just email plus Gmail or Slack connection - and Slack gets you more credits than Gmail. What's inside on the free tier: Claude Opus 4.8, Sonnet 4.6, GPT 5.5 and other current models. Limits are generous - not the usual "10 messages and you're out" situation. Real usage out of the box. Here's the exact flow: → Go to sauna(.)ai → Register with email → Connect Gmail or Slack when prompted - Slack gives more credits → Go to dashboard and start using 3 minutes. $100 in credits. Flagship models. No card. These promos disappear fast. Bookmark this post so you don't lose the link when it's gone.
English
AMD CEO LISA SU HELD A MINI PC ON STAGE THAT RUNS A 235B MODEL AND REPLACES YOUR $440/MONTH AI STACK
amd's ryzen ai max+ 395 is the first x86 chip that runs a 200 billion parameter model on one piece of silicon. cpu and gpu share 128gb of unified memory, no separate graphics card needed
the gmktec evo-x2 runs qwen3 235b fully, deepseek v3 comfortably and llama 3.3 70b with headroom. on linux you get 110gb of usable vram out of 128gb
amd claimed the chip beat an nvidia rtx 5080 by more than 3x on deepseek r1 inference. a lunchbox sized pc outrunning a $1,000 discrete gpu on a real ai workload
a heavy ai user pays $200 for claude code max, $200 for chatgpt pro, $20 for cursor and $20 for gemini. that's $5,280 a year and the box pays itself off in 9 to 10 months
install ollama, pull the model, point claude code at localhost. same interface, nothing leaves the machine, nothing costs per request
bookmark this and read the article below
starmex@starmexxx
English
@OpenRouter I wonder if this can be done with a combination of small local models
English
Introducing the Fusion API, the smartest compound model in the market.
Fusion achieves Fable-level intelligence at half the price.
How it works 👇
English
@simplifyinAI Athena code uses Hermes style memory to keep you agents contextualized
github.com/luckeyfaraday/…
English
Tencent just open-sourced Hy-Memory.
A memory plugin that gives Al agents real long-term memory using a 6-layer framework with dual reasoning.
→ System1: fast pattern matching for instant recall
→ System2: deep reasoning for complex memory retrieval
→ 35% reduction in token usage
→ 70% less memory bloat over time
Most agents forget everything between sessions. This fixes that. Works for long-running collaborative Al agents that need persistent context.
100% Open Source.
English
@0xSero Hermes memory is pretty good
It’s what I use for Athena
English
Every single time I see one of these I get excited until I read the words memory system.
Every single memory system I have tried is garbage, including the ones I’ve built.
Simplifying AI@simplifyinAI
Tencent just open-sourced Hy-Memory. A memory plugin that gives Al agents real long-term memory using a 6-layer framework with dual reasoning. → System1: fast pattern matching for instant recall → System2: deep reasoning for complex memory retrieval → 35% reduction in token usage → 70% less memory bloat over time Most agents forget everything between sessions. This fixes that. Works for long-running collaborative Al agents that need persistent context. 100% Open Source.
English
Luckey Faraday me-retweet
@luckeyfaraday @0xSero I do have another PC with a 4090 I'm wondering if I can do this even though they're not on the same device over a server I'm sure I can I just got to figure out the specifics
English
You can create your own mythos using fusion
OpenRouter has created a way to take all your favorite models and fuse them into one super smart model
If you're feeling sad after losing Fable, go ahead and give this a shot
OpenRouter@OpenRouter
Introducing the Fusion API, the smartest compound model in the market. Fusion achieves Fable-level intelligence at half the price. How it works 👇
English
@0xSero This is a great idea. Even without the hardware yet, it’s smart to download open models now while you still can. You can always run them later when you have the setup.
English
How life would be if the world weren’t run by some very questionable people:
English
Luckey Faraday me-retweet
