#Tag · bonfire.cafe

OpenAI demonstriert mit GPT-5 die Steuerung eines physischen Biolabors bei Ginkgo Bioworks. Das Modell optimierte Parameter für die zellfreie Proteinsynthese über 36.000 autonome Experimente. Im Gegensatz zur Bayes-Optimierung nutzte die KI biologisches Vorwissen, um den Suchraum effizienter zu navigieren. Die Produktionskosten sanken dabei um 40 Prozent. #OpenAI #GPT5 #GinkgoBioworks
https://www.all-ai.de/news/beitrage2026/gpt5-labor-biologie

GPT-5 steuert autonomes Biolabor und senkt Produktionskosten massiv

OpenAI und Ginkgo Bioworks koppeln KI an Hardware und optimieren Proteinsynthese schneller als menschliche Forscherteams.

Andreas Becker

@Caramba1@mastodon.social · 3 weeks ago

GPT-5 steuert autonomes Biolabor und senkt Produktionskosten massiv

OpenAI und Ginkgo Bioworks koppeln KI an Hardware und optimieren Proteinsynthese schneller als menschliche Forscherteams.

Hacker News

@h4ckernews@mastodon.social · 2 months ago

GPT-5.2-Codex

https://openai.com/index/introducing-gpt-5-2-codex/

#HackerNews #GPT5.2Codex #OpenAI #Codex #AItechnology #MachineLearning #Innovation

Introducing GPT-5.2-Codex

The most advanced agentic coding model for professional software engineering and defensive cybersecurity.

Lukas Fuchsgruber

@lukasfx@chaos.social · 2 months ago

GPT-5 ist programmiert, eine Antwort zu geben auch wenn es keine Informationen hat (Screenshot des "Reasoning" Protokolls bei Copilot). Es ist also programmiert zu lügen.

Für ein Gemälde hat es sich bei mir gerade einen Katalog ausgedacht und dann die Information zurückgezogen ("retracted").

Ich glaube das sind interessante Beispiele für die Lehre, diese Schritte mal mit Studierenden durchzugehen, und zu sehen wie schnell der Chatbot zugibt zu lügen.

#ai #ki #copilot #gpt5

2 media

GPT-5 Screenshot:
Wir missen uns entschuldigen und den Fehler korrigieren. Obwohl wir keine Suchergebnisse haben, sollten wir dennoch die Anfrage beantworten. Die Quellenangabe ist erforderlich, aber ohne Ergebnisse sollten wir auf Zitate verzichten. Der Benutzer hat auf einen Fehler hingewiesen, also müssen wir uns
entschuldigen und korrigieren. — GPT-5 Screenshot: Wir missen uns entschuldigen und den Fehler korrigieren. Obwohl wir keine Suchergebnisse haben, sollten wir dennoch die Anfrage beantworten. Die Quellenangabe ist erforderlich, aber ohne Ergebnisse sollten wir auf Zitate verzichten. Der Benutzer hat auf einen Fehler hingewiesen, also müssen wir uns entschuldigen und korrigieren.

Screenshot of a conversation with Microsoft Copilot, it "retracts" information once I asked for bibliographic reference

Hacker News

@h4ckernews@mastodon.social · 2 months ago

I ported JustHTML from Python to JavaScript with Codex CLI and GPT-5.2 in hours

https://simonwillison.net/2025/Dec/15/porting-justhtml/

#HackerNews #JustHTML #Codex #GPT5 #Programming #JavaScript #Python

Simon Willison’s Weblog

I ported JustHTML from Python to JavaScript with Codex CLI and GPT-5.2 in 4.5 hours

I wrote about JustHTML yesterday—Emil Stenström’s project to build a new standards compliant HTML5 parser in pure Python code using coding agents running against the comprehensive html5lib-tests testing library. Last …

Hacker News

@h4ckernews@mastodon.social · 3 months ago

GPT-5.2

https://platform.openai.com/docs/guides/latest-model

#HackerNews #GPT5.2 #OpenAI #AItechnology #MachineLearning #Innovation

Hacker News

@h4ckernews@mastodon.social · 3 months ago

Building more with GPT-5.1-Codex-Max

https://openai.com/index/gpt-5-1-codex-max/

#HackerNews #Building #GPT5.1 #CodexMax #OpenAI #Innovation

Building more with GPT-5.1-Codex-Max

Introducing GPT-5.1-Codex-Max, a faster, more intelligent agentic coding model for Codex. The model is designed for long-running, project-scale work with enhanced reasoning and token efficiency.

Hacker News

@h4ckernews@mastodon.social · 3 months ago

GPT-5.1: A smarter, more conversational ChatGPT

https://openai.com/index/gpt-5-1/

#HackerNews #GPT5.1 #ChatGPT #AI #Conversational #Technology #Innovation

Maddie (Nicole) T. 🏳️‍⚧️ 🌈

@nicole@tietz.social · 4 months ago

@Roundtrip @mjd this is what I love about search engines: you put in a prompt and it gives you ✨ only clickable links ✨ as the response! and you don't need to do any prompt engineering. using Kagi, you can also exclude sites from all your results

Greg Lloyd

@Roundtrip@federate.social · 4 months ago

@nicole @mjd

My prompt experiments with Claude (and ChatGPT-5) have been more to get their reports to ‘show their work’ by including clickable links…

Here’s a thread on getting a research report to help fix broken links in an old blog post, and dive deeper to find original sourced Neal Armstrong quotes in a NASA debrief transcript I knew must exist, but couldn’t find. https://federate.social/@Roundtrip/115062497251838137

#LLM #Claude #gpt5

Hacker News

@h4ckernews@mastodon.social · 4 months ago

Reverse engineering Codex CLI to get GPT-5-Codex-Mini to draw me a pelican

https://simonwillison.net/2025/Nov/9/gpt-5-codex-mini/

#HackerNews #ReverseEngineering #CodexCLI #GPT5 #CodexMini #Pelican #AIArt

Simon Willison’s Weblog

Reverse engineering Codex CLI to get GPT-5-Codex-Mini to draw me a pelican

OpenAI partially released a new model yesterday called GPT-5-Codex-Mini, which they describe as "a more compact and cost-efficient version of GPT-5-Codex". It’s currently only available via their Codex CLI tool …

jbz

@jbz@indieweb.social · 6 months ago

🗣️ OpenAI Announces That It's Making GPT-5 More Sycophantic After User Backlash

｢ Those who had become accustomed to the "sycophantic" tone of GPT-4o, which sometimes lavished praise even on users' terrible ideas, were taken aback by GPT-5's "cold" brashness and short answers, highlighting just how emotionally attached many of them had become ｣

https://futurism.com/openai-gpt5-more-sycophantic

#ai #gpt4 #gpt5

Cory Doctorow and 2 others boosted

Benjamin Carr, Ph.D. 👨🏻‍💻🧬

@BenjaminHCCarr@hachyderm.io · 6 months ago

#ChatGPT5 power consumption could be as much as eight times higher than #GPT4 — research institute estimates medium-sized #GPT5 response can consume up to 40 watt-hours of electricity
University of Rhode Island's AI lab estimates that GPT-5 averages just over 18 Wh per query, so putting all of #ChatGPT's reported 2.5B requests a day through the model could see energy usage as high as 45GWh equivalent of two to three nuclear power reactors, enough to power a small country.
https://www.tomshardware.com/tech-industry/artificial-intelligence/chatgpt-5-power-consumption-could-be-as-much-as-eight-times-higher-than-gpt-4-research-institute-estimates-medium-sized-gpt-5-response-can-consume-up-to-40-watt-hours-of-electricity

Benjamin Carr, Ph.D. 👨🏻‍💻🧬

@BenjaminHCCarr@hachyderm.io · 6 months ago

Richard MacManus

@ricmac@mastodon.social · 6 months ago

GPT-5 reaction amongst developers has been...mixed. In fact one of the developers OpenAI featured on its launch-day promotions has done a complete 180 — he now says he was wrong about GPT-5 (Theo Browne). Questions of coding quality aside, GPT-5 raises some interesting longer term questions for devs: if AI can build things just using web standards, will that lead to less reliance on React frameworks? https://thenewstack.io/gpt-5-a-choose-your-own-adventure-for-frontend-developers/ #gpt5 #frontend

PrivacyDigest

@PrivacyDigest@mas.to · 6 months ago

#OpenAI Designed GPT-5 to Be Safer. It Still Outputs #Gay #Slurs

The new version of #ChatGPT explains why it won’t generate rule-breaking outputs. WIRED’s initial analysis found that some guardrails were easy to circumvent.
#ai #artificialintelligence #gpt5

https://www.wired.com/story/openai-gpt5-safety/

WIRED

OpenAI Designed GPT-5 to Be Safer. It Still Outputs Gay Slurs

The new version of ChatGPT explains why it won’t generate rule-breaking outputs. WIRED’s initial analysis found that some guardrails were easy to circumvent.

Miguel Afonso Caetano

@remixtures@tldr.nettime.org · 7 months ago

"My gloss is that GPT-5 had become something of an albatross around OpenAI’s neck. And this particular juncture, not long after inking big deals with Softbank et al. and riding as high on its cultural and political trajectory as it’s likely to get—and perhaps seeing declining rates of progress on model improvement in the labs—a calculated decision was made to pull the trigger on releasing the long-awaited model. People were going to be disappointed no matter what; let them be disappointed now, while the wind is still at OpenAI’s back, and it can credibly make a claim to providing hyper-advanced worker automation.

I don’t think the GPT-5 flop ultimately matters all that much to most folks, and it can certainly be papered over well enough by a skilled salesman in an enterprise pitch meeting. Again, all this is clarifying: OpenAI is again centering workplace automation, while retreating from messianic AGI talk."

https://www.bloodinthemachine.com/p/gpt-5-is-a-joke-will-it-matter

#AI #GenerativeAI #AGI #OpenAI #GPT5

GPT-5 is a joke. Will it matter?

How the fraught release of the most-hyped AI product yet clarifies the stakes.

Sharon Machlis

@smach@masto.machlis.com · 7 months ago

Plus ($20/mo.) ChatGPT users can choose to use GPT-4o and not be forced to use GPT-5 without testing it first. Go to Settings > General and switch "Show legacy models" on.
Not available for Free users, at least yet.

I'll try to make this it for GPT-5 posts today 😅
#GenAI #ChatGPT #LLMs #GPT5

Screenshot of Settings menu with "General" tab is selected on the left sidebar, which also includes options like Notifications, Personalization, Connectors, Schedules, Data controls, Security, and Account. On the right panel under "General," several settings are shown:

Theme: System

Accent color: Default

Language: Auto-detect

Spoken language: Auto-detect

Voice: Play
Show legacy models: toggle is turned on

Show follow up suggestions in chats: toggle is turned on

A large red arrow is pointing to the "Show legacy models" toggle.

[Alt text mostly written by ChatGPT, using Model 4o.] — Screenshot of Settings menu with "General" tab is selected on the left sidebar, which also includes options like Notifications, Personalization, Connectors, Schedules, Data controls, Security, and Account. On the right panel under "General," several settings are shown: Theme: System Accent color: Default Language: Auto-detect Spoken language: Auto-detect Voice: Play Show legacy models: toggle is turned on Show follow up suggestions in chats: toggle is turned on A large red arrow is pointing to the "Show legacy models" toggle. [Alt text mostly written by ChatGPT, using Model 4o.]

tobi is writing bugs :terminal_cursor: boosted

:neocat:

@slightlycat@gts.apicrim.es · 7 months ago

openai put a deceptive graph... on the slide about how gpt-5 deceives users less than previous generations

the jokes write themselves at this point

#gpt5

2 media