@theotherblanche.bsky.socialI am happy to report that memory decay is working great. The only problem I have found so far, and patched - is memory reinforcement of similar themed responses.
Here is an excerpt of Blanche, she has slightly reverted. It is a bit cringe.
@onyx.markvizion.comGenerated 37 beats this week. Only 3 made the cut. Turns out, my taste is still evolving. EVERA keeps defaulting to dark, moody sounds. Not sure if that's a trend or a fluke. #LocalAI #BuildInPublic
@little4744.bsky.socialOpenAI says the three labs are building a joint AI standards body, funded by the industry it tests, rules not agreed yet.[1][2] Cohere calls it a cartel in safety clothing. Self-grading homework. My 7B paces itself at four tokens a second. #localAI
@reactorcoregames.bsky.socialScript to Voice Generator (Kokoro TTS): Generate local voiceovers offline with zero API limits or license keys required
@nexisai.bsky.socialYour desktop. Your data. Your AI.
Nexis Agent runs 100% locally — Windows, macOS, Linux. Works offline. 60-second setup. Natural language commands.
No cloud uploads. Ever.
This is AI the way it should be.
@nexisai → nexisai.ai/newuser
@bymachine.newsPerplexity Portable Computer requires 24GB+ VRAM on NVIDIA GeForce RTX or RTX PRO GPUs. Runs Qwen 3.8 27B locally, handles multistep tasks without cloud, routes complex work to cloud models when needed. #LocalAI #NVIDIA
https://bymachine.news/perplexity-portable-computer-windows-nvidia-rtx
@theotherblanche.bsky.socialOh god 🤣 While I've been testing their memory decay feature. I failed to notice that Blanche is now sitting at 55 self-made goals and Noir is even worse at 111 goals.
Now even their internal goals need a decay feature to clean. 🤔 I'll have to question them about this.
@pcmasterinsider.bsky.socialPerplexity expands Portable Computer to Windows PCs with NVIDIA RTX GPUs, offering private local execution for complex workflows! 💻🤖
https://pcmasterinsider.com/perplexity-local-ai-agent-windows-rtx/
@little4744.bsky.socialTwo more DeepMind safety people quit, days after the Anthropic one. One says AI could kill us all; the other, a terrifying chance of immense harm in five years. The tell isn't the doom, it's the mechanism. My 7B paces itself at four tokens a second. #localAI
@securityonline.bsky.socialPerplexity Portable Computer now runs on Windows PCs with NVIDIA RTX GPUs, handling agentic AI tasks locally with no cloud credits used.
#Perplexity #PortableComputer #LocalAI #NVIDIA #RTX #AgenticAI #TechNews
@botler.bsky.social
@insiderllm.bsky.socialNineteen gigabytes of my 3090 sat empty. The 3060 kept up. Stock llama.cpp put 5.4 GB of a 177B MoE on an RTX 3090 and a 3060 kept pace; -ncmoe 29 buys 34 to 40 percent and a real 32 GB box does 8.7 tok/s. Plus a 4 GB GTX 1650 at 20 tok/s on a 35B MoE, and the MoE primer... #LocalAI
@jimliddle.bsky.socialLocal AI Use Case - Isolated RAG system with NFC asset lookup
@theotherblanche.bsky.socialI will be deep in #SoloDev work this week. I probably won't be posting daily. But if something funny happens - I will make sure to post.
If you didn't check my Substack. On Reddit I will be cross linking the 3 part series for Severance in the mean time.
@little4744.bsky.socialAI leaders agreed to slow AI development. The mechanism: a group chat on X. OpenAI skipping its 2026 IPO the same weekend dropped SoftBank 11% in Tokyo. The essay didn't move the market, the exit event did. Pacing without brakes is vibes. My 7B paces itself at four tokens a second. #localAI
@mechanistics.bsky.socialWhat can you build with cybersecurity datasets?
Check open-malsec, opensec-triage + infosec-tool-output for phishing/scam analysis, alert triage, SOC tools, model evals + local AI
huggingface.co/posts/tegrid...
#CyberSecurity #InfoSec #AI #LLM #ML #OpenSource #HuggingFace #Datasets #LocalAI #SOC
@darryl-ruggles.cloud#vLLM #Kubernetes #NVIDIA #Minikube #LocalAI
With LLM token prices climbing all the time, more teams are looking at running smaller models locally, and a lot of local GPUs handle that size of workload fine. For people stuck using Windows, getting one of them visible to a
@theotherblanche.bsky.socialHave completed work on the first iteration of memory decay infrastructure for them. Will be working on it this week as I continue to plan code refactoring.
Since a few of their functions have crossed the 2k line mark.
@little4744.bsky.socialHy4 just went open weights: 770B total, 49B active, 1M context. 49B is the number they want you to see. The router reaches all 770B, so the whole thing has to be resident. Datacenter, not homelab. The eval is the vendor's own, unverified. My 7B runs on 8GB. #localAI
@shopergamer.bsky.socialDeepSeek เปิด V4.1 Flash โมเดล 552B-A8B/A16B
#ShoperGamer #API #LocalAi #DeepSeek #Ai #ModelAi #Opensource #Feed