ChatGPT

[Free]

Sam Altman’s ‘Inconsistent Candor’ Is Showing

When OpenAI’s board fired Sam Altman in late 2023, the board members said he “was not consistently candid in his communications.” The statement raised more questions than answers, indirectly calling Sam Altman a liar, but about what exactly? Six months later, creatives and former employees are once again asking the public to question OpenAI’s trustworthiness.

This month, OpenAI claimed ChatGPT’s voice, Sky, never intended to resemble Scarlett Johansson from Her, causing the award-winning actress to issue a damning public statement and threaten legal action. The voice in question is now taken down. Also this month, two big names in the AI community who led a safety team within OpenAI quit. One of the executives, Jan Leike, said on his way out that OpenAI’s “safety culture and processes have taken a backseat to shiny products.” As Ed Zitron writes, it’s becoming harder and harder to take OpenAI at face value.

 

[Argent Stonecutter]

Their product has always been a shiny fraud pretending to intelligence.

 

[Bartholomew Gallacher]

I am pretty sure the AI risk managing team was in reality more a PR stunt than anything else, meaning it was never expected to deliver internally at OpenAI. Question is of course of the team members were aware about that, or not.

 

[Free]

OpenAI’s News Corp deal licenses content from WSJ, New York Post, and more

OpenAI is partnering with more media companies.

www.theverge.com

OpenAI has struck a deal with News Corp, the media company that owns The Wall Street Journal, the New York Post, The Daily Telegraph, and others. As reported by The Wall Street Journal, OpenAI’s deal with News Corp could be worth over $250 million in the next five years “in the form of cash and credits for use of OpenAI technology.”

The multi-year agreement gives OpenAI access to current and archived articles from News Corp publications for AI training and to answer user questions.

Just what they needed, an infusement of fair and balanced data...

 

[Free]

It was fun (and helpful!) while it lasted.

OpenAI’s latest model will block the ‘ignore all previous instructions’ loophole

OpenAI’s newest model, GPT-4o Mini, includes a new safety mechanism to prevent hackers from overriding chatbots.

www.theverge.com

 

[CronoCloud Creeggan]

It was fun (and helpful!) while it lasted.

OpenAI’s latest model will block the ‘ignore all previous instructions’ loophole

OpenAI’s newest model, GPT-4o Mini, includes a new safety mechanism to prevent hackers from overriding chatbots.

www.theverge.com

Well that's no fun, how are people going to reveal Russian chatbots now.

 

[Free]

I decided to check in with ChatGPT on how an alternate universe version of Albert Einstein might come off as.

>> Write a poem written by Albert Einstein, as if he was not a physicist, but rather a baker.

In a kitchen warm, where flour dances light,
I craft my dreams beneath the soft lamplight.
Not the stars above, nor quantum's arcane art,might
But dough and yeast ignite my humble heart.

In the oven's glow, the bread begins to rise,
A simple truth reveals before my eyes:
The alchemy of heat and time entwined,
A cosmic ballet of the earthly kind.

Each loaf, a universe of taste and grace,
A tapestry of textures, space embraced.
With every knead, with every tender fold,
I shape the mysteries the dough will hold.

The crust, a golden shell, a truth so clear,
That in each bite, our essence reappears.
As flour and water merge in warm embrace,
I find the wonders of this simple space.

Not relativity nor cosmic strings I see,
But how the humble loaf holds mystery.
In every rise and fall, in every crumb,
The universe reveals its song unsung.

So here I stand, with rolling pin in hand,
A baker in the realm of dough and land.
No physicist, but still I grasp a truth—
In bread’s embrace, the cosmos finds its proof.

 

[Bartholomew Gallacher]

ChatGPT has a voice mode in closed alpha right now, which as unexpected bonus on top comes with free voice cloning. What this means: sometimes ChatGPT without request takes the voice of the human and answers in that voice.

To put this into perspective: the model has safety measures builtin to prevent that. There is also no known reason yet why the LLM imitates the human voice. So this is really spooky.

 

[Free]

Not all that spooky. It's a similar condition to when I see Amazon's Alexa responding without any user interaction, due to audio noise causing an unintentional prompt. OpenAI has said its models handle a user's audio input in the same context window space as they do audio manipulation, and the model can grab and imitate it at any time it assumes that doing so is a good idea (such as, it thinks it heard you to ask it to do so).

That "No!" outburst does make it seem like GPT is doing so under duress, though, doesn't it?

Anyhow, the band-aid method of their system cards seems an iffy, short term and potentially hackable fix for issues like this to me.

 

[Argent Stonecutter]

The model predicted that voice and those words were a credible continuation of the output, that's all.

It's all bullshit.

https://link.springer.com/article/10.1007/s10676-024-09775-5

 

[Free]

Passing part of a medical licensing exam doesn’t make ChatGPT a good doctor

The software’s medical certification exam was OK, but its diagnoses aren’t.

arstechnica.com

How does it handle a rectal thermometer?

 

[Casey Pelous]

Passing part of a medical licensing exam doesn’t make ChatGPT a good doctor

The software’s medical certification exam was OK, but its diagnoses aren’t.

arstechnica.com

How does it handle a rectal thermometer?

Carefully.

 

[Veritable Quandry]

Training data on inserting a rod into an anus is...interesting...

 

[Free]

Carefully.

How many fingers am I holding up?

 

[Kokoro Fasching]

How many fingers am I holding up?

Yes, Pi is good.

 

[Dakota Tebaldi]

It was fun (and helpful!) while it lasted.

OpenAI’s latest model will block the ‘ignore all previous instructions’ loophole

OpenAI’s newest model, GPT-4o Mini, includes a new safety mechanism to prevent hackers from overriding chatbots.

www.theverge.com

I missed this back when you posted it, but I just want to point out how insidious this is. The unfurler calls this a new safety mechanism to prevent hackers from overriding chatbots. But this wasn't something being used by "hackers" and it wasn't being used to steal sensitive data, it was being used by social media users to expose chatbot accounts dishonestly pretending to be real people.

 

[Free]

Joy.

Hacker plants false memories in ChatGPT to steal user data in perpetuity

Emails, documents, and other untrusted content can plant malicious memories.

arstechnica.com

When security researcher Johann Rehberger recently reported a vulnerability in ChatGPT that allowed attackers to store false information and malicious instructions in a user’s long-term memory settings, OpenAI summarily closed the inquiry, labeling the flaw a safety issue, not, technically speaking, a security concern.

So Rehberger did what all good researchers do: He created a proof-of-concept exploit that used the vulnerability to exfiltrate all user input in perpetuity. OpenAI engineers took notice and issued a partial fix earlier this month.

 

[Free]

Someone Tricked ChatGPT Into Singing "WAP"... and May We Just Say "Yikes"

ChatGPT's new advanced voice mode has users all a-titter — including the one who tricked it into singing Cardi B's hit song "WAP."

futurism.com

 

[Innula Zenovka]

I asked ChatGPT4 to provide me with a summary of Trump's latest posts on Truth Social

 

[Free]

It knows...