cedric (@cedric_chee)

ChatGPT-5.3 Instant에 대한 초기 사용기: ChatGPT-4o의 공백을 채울 가능성이 있으며 일상 대화에서 체감되는 개선이 있음. 더 직접적이고 방어적이지 않으며 공감적이고 건조하지 않고 '살아있는' 느낌(소울)이 난다는 평을 포함. 안전성 평가(harmless evals)도 일부 언급됨.

https://x.com/cedric_chee/status/2029055713348403593

#chatgpt #chatgpt5.3 #llm #modelrelease

Why Users Are Rallying to #Keep4o: The Social Backlash Against OpenAI Retiring ChatGPT 4o

To users, ChatGPT-4o was much more than an AI model... to many, it was a friend, a companion, and for some, even a lover..
We explore why ChatGPT-4o became so beloved, what motivated OpenAI's move to GPT-5, & how this conflict highlights tensions between rapid AI innovation and user experience.

#AI #OpenAI #Tech #ChatGPT #ChatGPT4o #ChatGPT5 #AIUpgrade #Technology #AIModel #LLM #GPT #ManiaAI #ManiaTech #GenerativeAI

https://maniainc.com/technology/why-users-are-rallying-to-keep4o-the-social-backlash-against-openai-retiring-chatgpt-4o/

"AI can write you excellent imitative verse that sounds Elizabethan, it cannot write you Shakespeare."

— Ben Affleck

This article is an interesting reprise of a rebuttal against AI, however I wonder whether it was written using a chatbot it's so clunky, stylistically repetitive, and strangely phrased. Ironic since the actor takes apart and destroys the idea that AI can replace the creativity of actors or filmmakers. Worth reading.

#ai #chatGPT #writer #author #writersOfMastodon #writingCommunity #film #filmmaking #affleck #llm #chatGPT5 #chatGPT6 https://www.buzzfeed.com/stephaniesoteriou/ben-affleck-viral-smart-ai-impact-hollywood-reacts?utm_source=flipboard&utm_content=user/Buzzfeed

Erdos 281 solved with ChatGPT 5.2 Pro

https://twitter.com/neelsomani/status/2012695714187325745

#HackerNews #Erdos281 #ChatGPT5.2 #Pro #AIResearch #ProblemSolving #Mathematics

Дрейф, потеря контекста и «уверенная чушь»: протокол восстановления SDX-S

LLM умеют многое, но иногда ломаются так, что виноватым выглядит пользователь: контекст уезжает, инструкции исчезают, инструмент падает, а модель продолжает говорить уверенно, как будто всё нормально. Мы смотрим на это не как на “плохой ответ”, а как на деградацию состояния диалога . Если не поймать момент, по цепочке шагов и становится всё убедительнее. Мы собрали процедуру SDX-S: триггеры → диагностика причины → восстановление → критерии возврата . Ниже: состояния, “дашборд” и два кейса, где это реально спасает.

https://habr.com/ru/articles/985334/

#сезон_ии_в_разработке #llmjs #chatgpt5 #prompt_engineering #guardrails #hallucinationsinai #observability #finite_state_machine #tool_use #reliability

a report on using web based LLMs for fluffy stuff...
> I'd estimate my direct use of ChatGPT at 2-6 hrs, 3-4 days a week ... therefore, most ChatGPT users spend much less time on it. https://write.as/3gmbckphg4wjpwj4.md

#LLMs #ChatGPT #ChatGPT5 #Claude #Lint

GPT-5.2登場能打贏Gemini 3嗎？3大新模式差異、開放資格一次看

#7365 Report: The 2000W camera on these AI Smart Glasses inadvertently overcooked Kevin's crumpet during focus tests. Imaging was 1080P compliant. Now €482.64 (was €509.63). Battery (400Mah) held charge. Rated 4.735★.
#SmartGlasses #Tech

#chatgpt5.2 #TheFindLab #tested

https://s.click.aliexpress.com/e/_c3zogZxX