ken mercado

@photobiogenesis I had a massive and continued improvement in GI issues after psylium husk. It's a miracle fiber.

More research papers have been published supporting my assertion that higher intake of soluble fiber is the most important factor in reducing zonulin, the factor that drive intestinal permeability and "leaky gut" This one was published just this month sciencedirect.com/science/articl… They found that incorporating 30g of dietary fiber for one month was enough to lower zonulin, circulating endotoxin (LPS), and thyroid antibodies (anti-TPO) The subjects also lost weight, likely as a result of greater satiation and lower caloric intake

Anyone know the best AI to find best LA apartments/houses based on criteria? GPT is so shitty at this and is simply looking at text descriptions

I fucking love Capitalism

@badlogicgames Same experience.

i have real trust issues with opus. it's extremely weird. any time it suggest a thing, i switch to gpt and ask for a second opinion. in +60% of cases, gpt finds additional things opus just missed. (not a gpt shill, love opus, but it's complicated)

@DrCamRx When I was 18, I couldn’t even do 3 pull ups lol

Unpopular Opinion: you are NOT as old as you feel. You are as old as you function. Your strength & speed IS your age: 20s: 20+ pull-ups & <60s 400m run 30s: 10-19 pull-ups & <75s 400m 40s: 5-9 pull-ups & <90s 400m 50s: 1-4 pull-ups & <120s 400m OLD: 0 pull-ups & 120+s 400m

Ts is why Facebook has remained the dominant probabilistic forecasting ad engine on the planet for the past decade. Zucc will never cease to explore ways to absolute rape the human mind. Beautiful.

@dejavucoder set aut compact at 10-20% and use morphllm

man claude code has stopped doimg the clear context + auto accept edit when using 1M opus. it just shows auto accept edit now. I AM TOO AFRAID TO USE CLAUDE POST 100K TOKENS, PLEASE LET ME CLEAR CONTEXT AND BEGIN THE PLAN!

@elliotarledge Lol it’s more like .00001

less than 0.01% of the population understands how a transformer works

(What i do) You can write harnesses using bash script or python, and then call non-interactive `claude -p -bare` instances. For research use cases, I'm guessing you can write research subagent (claude -p) and then evaluate it's work with another subagent, then maybe some self-improving step where it can modify research prompt. The bash script would loop indefinitely. You'd have to constantly review to make sure it's doing what you want tho can use strategies from: Harness Design for Long-Running Apps — anthropic.com/engineering/ha… Effective Harnesses for Long-Running Agents — anthropic.com/engineering/ef… Multi-Agent Research System — anthropic.com/engineering/mu…

@shantanugoel u wrote the harness yourself?

what is the best tooling for 24-7 inference/agent-driven research? im trying factory but it stops and asks me questions even though i have 'auto' mode on. tbh i think this is an even bigger killer app than LLM chatbots. who else is out there doing it?

Thank you

@spiritbuun Autoresearch could have discovered if configured to research related stuff in game engine research

The funniest thing about TurboQuant is that not only is none of it new, every part of what we've implemented is very old. We are just rediscovering tricks video game developers had to use in the 90s. FWHT decorrelation → DCT in JPEG/MPEG decorrelation Lloyd-Max 16-centroid codebook → 256-color palette (VQ codebook) Norm + quantized unit vector → Quake's norm vector table 128-elm blocks with 4-bit indices → S3 texture compression ADC lookup table (VecInfer) → Dot-product LUTs for lighting Per-block scale factor → ADPCM audio (scale per block + quantized residuals)

@spiritbuun Autoresearch could have discovered it

@LuciferMornens @morphllm note it crashed my claude code at first had to ask claude to debug and fix

@fastquant0 @morphllm how do you use this in CC?

Claude compacts are now 50% faster while improving retention & accuracy. Thanks @morphllm

@LuciferMornens @morphllm claude help me setup morphllm.com

@ColtonOrtolf Enlicitide (oral version) coming in 1-2 years

Evolucumab is a miracle drug. When added to statins, it causes greater plaque regression by decreasing atheroma volume (-0.95% vs +0.05% on statins), increasing fibrous cap thickness (42.7 µm vs 21.5 µm), reducing maximum lipid arc (-57.5° vs -31.4°), decreasing the macrophage index (-3.35 vs -1.43), and lowering Lp(a) by 27%, unlike statins which may increase it.

We should all be on statins. What an incredible drug

@AnthropicAI So I'm guessing it can dramatically improve if dedicated RL is done

@AnthropicAI Opus optimizes prompts badly if tasked to do so. It adds way to specific rules to prompt when given several examples. It also tries to add conditions to prompt when it can alternatively outsource such conditions to code.

uhhh @AnthropicAI What if you do RL on building harnesses? anthropic.com/engineering/ha… Train the model to learn robust harness building strategies. Have the model bias towards concepts & tools that improve its performance on extremely long running tasks. Better harness building = much more likely to have maintained coherence in very very long running tasks as model would have learned efficient and robust harness building concepts and strategies.