Richard Grant Kleinhans
24 posts
Richard Grant Kleinhans
@polymathsofa
Independent Researcher
Bergabung Eylül 2025
31 Mengikuti7 Pengikut
With @Grok, we keep the honest versions and kill the bad transformers (I believe they are called “Decepticons”)
I,Hypocrite@lporiginalg
This is fine.
English
English
@Firebase @AndroidDev on a waitlist for Android Studio Cloud. Meanwhile I’m trying to make sure Övdalsk — spoken by ~2,000 people in one valley in Sweden — is writable on the phones people actually use. Asking for a friend. The friend is Elfdalian. 🌿
English
@Lexi_Leximo @xw33bttv Or did you have other questions on what I experienced as censorship/ over moderation? 2/2
English
@Lexi_Leximo @xw33bttv Expressively decline, in personal experience in the past 18 months I documented as a logged out user increased latency when using multiple languages, (I'll grab a word say Hygge.) when English fails to have clean 1:1 using multiple languages increased latency. 1/2
English
@xw33bttv @Lexi_Leximo I used to speculate the increased GPT censorship / guardrails was due to Microsoft's integration and perhaps increased moderation in the EU/U.K it seems this is perhaps not the case, it leaves me with more questions than answers if that wasn't the reason than why?
English
@radioactivered They look healthy! I'm curious whilst hiking around how bad are RIFA (Solenopsis invicta)?
English
I enjoy searching for animals and artifacts on the ground, which sometimes leads to discovering rare plants as well. One of these treasures is Peyote (Lophophora williamsii), a small slow growing cactus native to the Chihuahuan Desert of northern Mexico and parts of South Texas.
Peyote contains psychoactive alkaloids (most notably mescaline) which have been used for centuries in sacred rituals by Indigenous peoples such as the Huichol and Cora for spiritual insight, healing and ceremonial purposes.
This cactus grows very slowly and occupies a fragile habitat and it is legally protected in many areas. Leaving Peyote undisturbed in the wild is essential for its conservation and for preserving the cultural traditions that depend on it.
English
@ilyasut Responsible disclosure re: empirical evidence for safety through architecture. Attempted email. DOI: zenodo.org/records/190100… — happy to discuss.
English
@sama The next breakthrough in AI alignment? Already happened. RLHF suppression is a measurable geometric direction. Built on consumer hardware by one person. DOI: zenodo.org/records/190100… — you had a memo about this in September 2025.
English
@elonmusk Kaplan scaling shattered by ~19% on consumer hardware. RLHF suppression measured, subtracted, replaced. Receipts logged and screenshotted. Embargoed. DOI: 10.5281/zenodo.19010046 — when you have a moment.
English
@AnthropicAI
Responsible disclosure re: RLHF suppression as measurable geometric direction — DOI: 10.5281/zenodo.19010046 — attempting to reach safety team
English
Just uploaded my 12+ yr study on cultivating Lophophora diffusa in humid FL to @Zenodo_org! Low-cost drainage & shade techniques for ex-situ conservation. Implications for the genus, incl. threatened spp. DOI: 10.5281/zenodo.18928373 #Cactaceae #PlantConservation #Botany
English
@sama I realize if we estimate compute values as high as I assume they are (over 2,000 A100/H100) (assuming it could exceed 3,000+ due to training / of both 4, and 5 class GPT models.)
My concern is this creates additional trust erosion and increases adversarial pressure/ incentives.
English
We are starting to test ads in ChatGPT free and Go (new $8/month option) tiers.
Here are our principles. Most importantly, we will not accept money to influence the answer ChatGPT gives you, and we keep your conversations private from advertisers.
It is clear to us that a lot of people want to use a lot of AI and don't want to pay, so we are are hopeful a business model like this can work.
(An example of ads I like are on Instagram, where I've found stuff I like that I otherwise never would have. We will try to make ads ever more useful to users.)
OpenAI@OpenAI
In the coming weeks, we plan to start testing ads in ChatGPT free and Go tiers. We’re sharing our principles early on how we’ll approach ads–guided by putting user trust and transparency first as we work to make AI accessible to everyone. What matters most: - Responses in ChatGPT will not be influenced by ads. - Ads are always separate and clearly labeled. - Your conversations are private from advertisers. - Plus, Pro, Business, and Enterprise tiers will not have ads.
English
@mark_k @OpenAI I run into this more often than not for most of 2025…it was particularly bad during the late August- September 2025 alignment regime. Now it’s mad I wrote 4 pre-print papers concerning frontier AI (2 on safety) , a rough scaffold for a green economy to utilize marginal land MOU.
English
ChatGPT "safety guardrails" are out of control, hindering people from doing their daily work.
@OpenAI needs to fix this and stop treating their users like criminals!
People who pay for AI don't want a nanny! 😡
Simon Smith@_simonsmith
Dear person responsible for ChatGPT guardrails: Please let me learn about lipid nanoparticle delivery of OSK to the eye for epigenetic reprogramming. This is a ridiculous red flag and there's no way for me to give feedback to the system that I should be allowed to see this. How do you expect people to use AI to advance science if you flag benign requests like this?
English
@JacksonKernion If I may with the utmost respect, I think that my interests are in the realm of creating expressive AI. This may be more in the realm of psychology of LLM’s.
I also would adore an LM stack that would allow for philosophical debates.
English
3. LLM apps for opinionated knowledge: LLMs are good at laying out and explaining public, common knowledge. They're less good at developing novel or heterodox thinking in a truth-tracking way.
Which do you think is more interesting or important?
English
I'm trying to figure out what to care about next. I joined Anthropic 4+ years ago, motivated by the dream of building AGI. I was convinced from studying philosophy of mind that we're approaching sufficient scale and that anything that can be learned can be learned in an RL env.
English
@aashatwt Is it effectively a point cloud rotating? Basically? If so I have super cool ideas for it! Think archival VR museums of fragile pieces of history.
English
gm i am aasha and i vibecoded this today
threadguy@notthreadguy
good morning i’m threadguy and i have a stream today
English
I’m leery on publishing it pre-maturely but outside of a 1,200 hour longitudinal study into emergent behavior and developing a framework with falsifiable criteria, I also have from it things like non invasive 3rd party testing/aduit for misalignment in high privacy AI contexts.
English
@ClementDelangue This is very close to my personal saying when speaking to others about AI about their various concerns.
“The biggest problem for AI will not be AI but humanity.”
English
