We adapt to AI’s limitations, not vice versa

Linguistics Professor Emily Bender on AI
Computer History Museum interview

We very quickly learn you can only ask [a Magic 8-Ball] yes/no questions if you want to have a successful conversation…and we’re doing the same thing with chatbots. We are telling ourselves a story as we put an input into the chatbot so that we can use that story to make sense of what comes back, Dr Bender said in her talk with David Brock.

I knew a kid in 4th grade that used to keep a Magic 8-Ball in her bag and consult it for, I don’t know, who to sit with at lunch.

Using chatbots, we’re getting fluent in AI-prompting, like we learned to hone our Google searches, but the effect on us is pretty unconscious.

Dr Bender shows how we adapt our thinking to accommodate AI’s limitations. We’re adjusting our prompts – so therefore our thoughts – to fit what we believe AI can handle (a college professor friend told me he can’t get the post-ChatGPT students to come up with ideas without consulting their AI 8-Ball).

Dependency risks stunting our thinking. And then it’s easy to mistake the AI answers for AI understanding (we’re the one’s doing the understanding bit).

Is it a fair deal, to train ourselves to think in a way to yield better AI answers, and risk our own potential good sense? [37 min. watch] youtube.com

Just because something worked when it was new, doesn’t mean it won’t need attention down the road

Salt Typhoon and patching
Steve Gibson has thoughts about the reported breaches of telecom companies, including Charter Communications

Salt Typhoon is a Chinese-state-backed hacking/spying group that has infiltrated, at the least (these are the one’s that have been reported), many US telecom companies.

It’s speculated that the infiltrated phone records give the Chinese fairly complete information about the whereabouts of US government officials, business leaders, competitors, and no telecom is sure that the spying has ended.

But one thing about these telecom companies that really upsets Steve Gibson as he explains on his Security Now podcast, is that the compromised entry-points were routers that had been patched by the manufacturer Cisco years ago, but the telecoms did not apply these patches … [2 hr. listen] twit.tv

How much privacy remains possible?

Practical privacy

A friend told me that there is no privacy if we carry cell phones. Partly true.

I found this Electronic Frontier Foundation’s (EFF) guidance helpful in giving an overview of what’s happening with devices and software that compromise our privacy.

Yes, EFF admits to limitations in what we can do to prevent triangulating our locations if the devices are on and in our pockets, but Warren and Brandeis witnessed and foresaw a kind of intrusion that in their words, both belittles and perverts our lives. Belittles by exaggerating the importance of what may be trivial in the breadth of our lives. Perverts by being able to access and using select words or actions against us. Except in genuine criminal cases, who benefits by this kind of access and scrutiny?

EFF offers a guide of what we can do whether or not we carry Android or Apple phones with us … [10 min. read] eff.org

The 10-year moratorium to stay individual states’ regulatory power over AI was stopped loudly, bipartisanly in the Senate

States retain their rights to regulate AI

Speaking of privacy concerns and looking for a bit of accountability from Big Data, the federal government will not be able by law to limit states’ rights to represent their citizens’ interests concerning the AI-tech mammoths.

And loss of privacy and personal and business information are concerns of AI-chatbot use – the more data we put in, the more of a challenge it is for Google, Meta, Amazon, etc. to control.

Sure, the chatbot agents have a nice interface, but how sure are we about the code on which these various platforms are built? [4 min. read] arstechnica.com

Lessening the impact of mistakes

To err is human
Minimizing employee risks

One of 2024’s most prominent breaches, the Change Healthcare cyberattack, was attributed to human risk. A low-level employee’s credentials were compromised through a phishing email, allowing the attackers to gain access to the network without multifactor authentication, enabling them to exfiltrate sensitive data and deploy ransomware.
Bryley partner Mimecast recommends steps to address and lessen human risk factors like limiting permissions (so if there’s an error, it can be contained to the extent of the responsible employee’s access) … [7 min. read] mimecast.com

^{Note: The section directly above is Bryley’s curated list of external stories. Bryley does not take credit for the content of these stories, nor does it endorse or imply an affiliation with the authors or publications in which they appear.}