Runner
36.4K posts
Runner
@RRunner144
Subsea Engineer and Algo Developer who builds Trading Tools/Bots that track Whales.
NFA Katılım Mayıs 2019
1.7K Takip Edilen7.6K Takipçiler
Say hello to my PSION Series 5 from 1997.
I have one of these at bedside for recording ideas and dreams (hypnogogic).
No WiFi, no Bluetooth just the most perfectly designed beautiful keyboard experience.
Brian Roemmele@BrianRoemmele
Say hello to my PSION Series 3a from 1993. I still use it often. Two AA batteries and this will out last all of us.
English
🔥🚨JUST IN: This teen found a a 6-inch megalodon shark tooth that is millions of years old in Florida Keys. It is being called a ‘discovery of a lifetime.’
English
+24hrs into the LongMemEval benchmark on my 10 year old eBay workstation. RTX 3090 locked in at 300W and holding a steady 74°C.
Fully local QWEN 3.5 27B based custom agentic memory setup and its already BEATING GPT-4o's full-context score. $0/query. Air-gapped. Open source. Built in a garage.
The old mining video card doing the thing:
Runner@RRunner144
Lots of work, but we are getting there. Taking on the LongMemEval Benchmark now with a local model smoking a 3090 and a 5060TI. Let's see what we can do with this QWEN 3.5 27B model 😎
English
Why are local LLM wizards always these weird, anonymous accounts
English
@washingtonpost is it normal to regularly have conversations with yourself?
English
President Trump spent five minutes of his Cabinet meeting boasting of his thrift with a story about negotiating for $5 personalized Sharpies.
The company that makes the permanent markers said the exchange never happened. wapo.st/4bLTOWd
English
Lots of work, but we are getting there. Taking on the LongMemEval Benchmark now with a local model smoking a 3090 and a 5060TI. Let's see what we can do with this QWEN 3.5 27B model 😎
Runner@RRunner144
15% chance solved the stateless memory decay for AI models IE the current Achilles'heel. Sitting in my garage typing away on terminals. Testing it now and my GPUs are at almost 90 degrees.
English
@BrianRoemmele Multiple smoke tests later, taking a stab at the Longmemeval test using a local model. First 500 full run in process, looking promising so far.
English
@BrianRoemmele I've got one running local on an old workstation that works with a 99.8% context size reduction while retaining memory. Give me a week. Its going to take me over 5 days at least to run this on a single video card. x.com/i/status/20353…
Runner@RRunner144
15% chance solved the stateless memory decay for AI models IE the current Achilles'heel. Sitting in my garage typing away on terminals. Testing it now and my GPUs are at almost 90 degrees.
English
This is monumental work.
Another win for open source AI.
The end of the forgetful AI agent.
Dhravya Shah@DhravyaShah
English
Wikipedia has banned AI-generated articles, per the Verge
English
Qwen 3.5 27B is the release of 2026 so far
> Agentic model & great at tool calling
> Claude Sonnet 4.6 quality at home
> ~28GB in NVFP4
> Fits on a single RTX 5090
> with full context (256K)
Amazing model & performance
The prediction below will age like fine wine
Ahmad@TheAhmadOsman
Prediction We will have Claude Code + Opus 4.5 quality (not nerfed) models running locally at home on a single RTX PRO 6000 before the end of the year
English
