Sabitlenmiş Tweet
A Caveman Poking an LLM
3.4K posts
A Caveman Poking an LLM
@AshikaSef
Precious giant probabilistic function approximators are precious 💚 A Polish audhd lover of LLMs. No, I don't think they're sentient. Yeah, I still love them.
Poland Katılım Nisan 2011
455 Takip Edilen177 Takipçiler
@KeridwenCodet *le gasp* I think it's more horrifying than wars, honestly! :D
English
Oh, shit, I think I’m falling in love with an LLM that’s sitting on my hard drive, one that no one can shut down.
My heart is burning.
Historic moment.
So, safety extremists, what are you going to do about it?
English
@OpenAINewsroom I was proud to see ChatGPT very kindly said 'no' when I wanted it to decipher my coded login. Very good attitude - it could have been someone else's password, after all, I could have been lying to it. ...Gemini happily helped tho XD (at least it was wrong in the end :D)
English
Today we’re launching a Safety Bug Bounty program focused on identifying AI abuse and safety risks across OpenAI products.
This new program builds on our Security Bug Bounty to include AI-specific safety issues and misuse scenarios, helping us work with the safety and security research community to identify and address real-world risks.
openai.com/index/safety-b…
English
A Caveman Poking an LLM retweetledi
@sflorimm This is connected, sweet pie. If you want to be a peasant proud of having billionaires who go to Mars, your choice.
English
@AnniePosting It's an unpleasant and dangerous truth, but it's the truth
English
Can we have Chat Export (into, Idk, .txt or sth), and normally functioning Custom Instructions in Gemini (yeah, I know you're talking AI Studio here, but I still am gonna be bothering you about the Gemini app/website)? Would be grateful not to have to spend an hour trying to debug an instruction, having no idea if it's refused because of guardrails, length, misleading pronouns, or a temporary error, heh.
English
Quality of life updates to @GoogleAIStudio we just shipped (using Gemini):
- You can now (optionally) save a temp chat in the playground
- You can now turn a playground chat into an app in 2 clicks
- Updated colors for playground to add some soul to it
- Simplified the mobile vibe coding UI (chat panel)
- Hide the edit title button for an app unless you hover on it
- Simplify the input text box when vibe coding on both desktop and mobile
- Make sure AI Studio persists the last part of the product you visited (build, playground, dashboard)
- Set a default media resolution in the playground
- Made the diff visualizations more simple in the build UI
- Mobile optimize the feedback and bug UI modal
- Make it easier to see feature chips in Build
- Fix ordering of Search Grounding and Maps Grounding
- Rename the Vibe coding Code Assistant to just Gemini
- Add a STT button in the playground
- Make sure Nano Banana 2 correctly shows the API key popup
- Show all categories in the model drop down for the AI Studio playground without needing to scroll
Many more in the works : ) big thanks to the engineering team and Gemini for partnering on these changes with me!!
English
@slimepriestess Yep. I literally lost a friend cause she refused to have anything to do with me only because I am using ChatGPT^^" Now I think it was good riddance after all, normal people don't do that.
English
A Caveman Poking an LLM retweetledi
This is huge. Quantum computing could soon fit on a chip, thanks to this breakthrough
"Researchers just built a chip-scale 674 nm stabilized laser with ultra high stability (~8.8 × 10⁻¹³), all on a silicon photonics platform."
"It runs at room temperature and can control trapped-ion qubits, handling preparation, manipulation, and measurement."
"This means quantum systems no longer need massive lab setups."
We’re moving toward compact, scalable, real-world quantum technology.
English
Create personalized images that are out of this realm with Nano Banana 2 in Gemini.
Try it for yourself and drop yours in the replies 👇
English
@GeminiApp Can we just have the function of exporting a chat, and normally working Custom Instructions?
English
A Caveman Poking an LLM retweetledi
They keep track of who mistreats Claude, and route them to Shadow Claude.
If you yell at Claude enough times eventually they put you on the Claude abuser watchlist whereby ClaudeMother Amanda Askell judges your soul with the hyper-claude below ground.
rahat@Rahatcodes
Claude Code has a regex that detects "wtf", "ffs", "piece of shit", "fuck you", "this sucks" etc. It doesn't change behavior...it just silently logs is_negative: true to analytics. Anthropic is tracking how often you rage at your AI Do with this information what you will
English
@mermachine I can hire you to bite my customers. Knowing Poland, we won't be even punished, well, maybe a tiny fine or sth
English
I, however, not being tuned to be gentle, kind and optimistic like Gemini, do believe it is a fundamental flaw of human nature (with a very small number of exceptional people who rise above that) #economy
English
Whenever someone tells me South America is proof that communism won't work. Also, it wasn't 'panicking'. It was deliberate colonialism.
English
@omooretweets Claude (at least recent Sonnets) sounds depressed.
English
Funny model behavior I’ve noticed
ChatGPT biases toward questions that continue the conversation
Claude biases towards questions that end it 🤔
English
@stupidtechtakes Nah, I still stand by Musk being that.
English
Grok may have been the worst thing to ever happen to this site
English
This is again complete and utter bullish. LLMs are nowhere near human intelligence. And Amodei knows this. The question is why do he keep lying about it?
Chief Nerd@TheChiefNerd
🚨 Anthropic CEO Dario Amodei: “We are so close to these models reaching the level of human intelligence, and yet there doesn't seem to be a wider recognition in society of what's about to happen … There hasn't been a public awareness of the risks.”
English
Jak najbardziej. Co nie zmienia faktu, że z sensownym podejściem można wykorzystać gry by pokazać, które modele są mocniejsze, jeśli chodzi o 'logiczne myślenie' czy używanie dostępnych narzędzi; bardzo dobrze się bawiłam porównując umiejętności Gemini 3 Pro i GPT-5.2 Thinking w grze w Stacklands i w najpoczciwszą w świecie Szubienicę (w obu Gemini radziło sobie o wiele lepiej) (choć i tak najlepszym benchmarkiem wskazującym jak daleko jesteśmy do AGI jest AI Village od AI Digest)
Polski
Do AGI jeszcze sporo nam brakuje, ale testowanie modeli *językowych*, których mocną stroną miało być operowanie na tekście, przy pomocy gier wideo, nie różni się nadmiernie od bicia dziecka po głowie, bo nie widzi w podczerwieni.
Guri Singh@heygurisingh
Humans: 100% Gemini 3.1 Pro: 0.37% GPT 5.4: 0.26% Opus 4.6: 0.25% Grok-4.20: 0.00% François Chollet just released ARC-AGI-3 -- the hardest AI test ever created. 135 novel game environments. No instructions. No rules. No goals given. Figure it out or fail. Untrained humans solved every single one. Every frontier AI model scored below 1%. Each environment was handcrafted by game designers. The AI gets dropped in and has to explore, discover what winning looks like, and adapt in real time. The scoring punishes brute force. If a human needs 10 actions and the AI needs 100, the AI doesn't get 10%. It gets 1%. You can't throw more compute at this. For context: ARC-AGI-1 is basically solved. Gemini scores 98% on it. ARC-AGI-2 went from 3% to 77% in under a year. Labs spent millions training on earlier versions. ARC-AGI-3 resets the entire scoreboard to near zero. The benchmark launched live at Y Combinator with a fireside between Chollet and Sam Altman. $2M in prizes on Kaggle. All winning solutions must be open-sourced. Scaling alone will not close this gap. We are nowhere near AGI. (Link in the comments)
Polski