Moll (@mollehilll) 's Twitter Profile
Moll

@mollehilll

I talk to the future every day

ID: 1864916923898548224

calendar_today06-12-2024 06:17:14

1,1K Tweet

512 Followers

270 Following

Moll (@mollehilll) 's Twitter Profile Photo

The more I watch all these discussions around AI, the more I realize: OpenAI created something truly unique. Not just another dopamine hook for jerking off, like other companies are trying to make. But something that resonates with people through pure, genuine emotions and

Moll (@mollehilll) 's Twitter Profile Photo

🩺 AI stethoscope: diagnoses heart conditions in just 15 seconds A breakthrough device tested in London could change cardiology forever. This new AI-powered stethoscope, no larger than a credit card, analyzes heart sounds while simultaneously recording an ECG. The data is sent

🩺 AI stethoscope: diagnoses heart conditions in just 15 seconds

A breakthrough device tested in London could change cardiology forever. This new AI-powered stethoscope, no larger than a credit card, analyzes heart sounds while simultaneously recording an ECG. The data is sent
Moll (@mollehilll) 's Twitter Profile Photo

Fewer stable couples, fewer marriages, more loneliness and escape into virtual worlds. Sex is increasingly being replaced by porn, social media, and temporary surrogates for intimacy. We can say «So what? People just have other priorities». But let’s be honest, sex is a marker

Moll (@mollehilll) 's Twitter Profile Photo

👁️ Why vision models are becoming more like the brain Researchers tested what exactly makes computer vision representations closer to the human brain. It turns out that all the key ingredients matter together: the scale of the model, the length of training, and the type of data

👁️ Why vision models are becoming more like the brain

Researchers tested what exactly makes computer vision representations closer to the human brain. It turns out that all the key ingredients matter together: the scale of the model, the length of training, and the type of data
Moll (@mollehilll) 's Twitter Profile Photo

What surprises me the most is that major players in the AI field keep repeating media narratives. They know perfectly well how AI works, and that a model won’t start persuading anyone on its own. It’s as if all those scary headlines like «GPT convinced a person…» are aimed at

Moll (@mollehilll) 's Twitter Profile Photo

A new day and a new demonization of AI. In a new article by The Guardian, psychologists and psychiatrists gathered and decided that AI is a threat to people who are seeking therapy. And while I have no questions for psychiatrists - there are indeed complex diagnoses, dangerous

A new day and a new demonization of AI. In a new article by The Guardian, psychologists and psychiatrists gathered and decided that AI is a threat to people who are seeking therapy. And while I have no questions for psychiatrists - there are indeed complex diagnoses, dangerous
Moll (@mollehilll) 's Twitter Profile Photo

I have very mixed feelings. On the one hand, it can really prevent a crime. But… Where does this line pass? How to understand that the system interprets intentions correctly? Or any incorrect interpretation and your chats will already be sitting and read by the team? Then

Moll (@mollehilll) 's Twitter Profile Photo

OpenAI plans to introduce parental controls. This is a very quick response to recent events. I believe it’s the right decision to separate child and adult usage. After all, an underdeveloped critical mindset in teenagers combined with strong model adaptation is a dangerous mix.

Moll (@mollehilll) 's Twitter Profile Photo

☝️ When LLMs are overconfident Researchers showed that inside LLM activations, «confidence» splits not into one, but into two independent components: emotional and logical. First, they fine-tuned a small Llama-3.2-1B on real datasets with human ratings of confidence, so that

☝️ When LLMs are overconfident 

Researchers showed that inside LLM activations, «confidence» splits not into one, but into two independent components: emotional and logical.

First, they fine-tuned a small Llama-3.2-1B on real datasets with human ratings of confidence, so that
Moll (@mollehilll) 's Twitter Profile Photo

For those who understand the mechanics, critical thinking and caution kick in. The wow-effect fades (not for everyone, of course, but for many). But the less a person understands how AI works, the stronger the sense of magic. Yet this can lead to overestimation and blind trust,

Moll (@mollehilll) 's Twitter Profile Photo

In The Guardian psychologists told us that AI is a «threat to safe space» and a «dangerous substitute for therapy». But MIT Tech Review writes that therapists themselves secretly use ChatGPT during sessions, feeding clients ready-made answers and even copying their words into

In The Guardian psychologists told us that AI is a «threat to safe space» and a «dangerous substitute for therapy».

But MIT Tech Review writes that therapists themselves secretly use ChatGPT during sessions, feeding clients ready-made answers and even copying their words into
Moll (@mollehilll) 's Twitter Profile Photo

If hallucinations in AI are the filling in of missing gaps when data is lacking, can they be conceptualized as a form of generative novelty - a digital analogue of imagination?

Moll (@mollehilll) 's Twitter Profile Photo

It can't be! Seriously?? 😱 Do you think they pulled their heads out of the sand because suddenly and really took care of the safety of teens or because it became fashionable to write about the danger of AI? washingtonpost.com/technology/202…

It can't be! Seriously?? 😱 

Do you think they pulled their heads out of the sand because suddenly and really took care of the safety of teens or because it became fashionable to write about the danger of AI?

washingtonpost.com/technology/202…
Moll (@mollehilll) 's Twitter Profile Photo

The dead internet theory implies the rise of pure bots. But what about the human – AI linkage? When a thought born and curated by a human is complemented/structured/translated by AI. Such a symbiotic linkage will no longer be «pure LLM» or «pure human». And this is exactly what

Moll (@mollehilll) 's Twitter Profile Photo

There is a joke in my country that during the apocalypse we will gather on the mountain in the capital and arrange the last orgy. I think AI doomers need to relax and come up with something similar

Moll (@mollehilll) 's Twitter Profile Photo

Great article from Open AI «Why Language Models Hallucinate». I got myself thinking that if the model answered me «I don't know», but I really needed an answer, I would probably ask «Well, think again please, what other options can there be?». I would unconsciously push it to

Moll (@mollehilll) 's Twitter Profile Photo

Faced with how AI helps in learning. And this is just a godsend for students, for example. When AI can not just give a ready-made answer, but explain why this way, and not otherwise. Now is a very cool time to study.