It means that was in its training. LLMs aren't aware of anything.
06.03.2026 19:17
👍 1
🔁 0
💬 0
📌 0
It means that was in its training. LLMs aren't aware of anything.
That's over 7 hours of downtime in one month.
Stripe has less than 5 minutes in a year.
Congratulations, Anthropic, on achieving one nines reliability in February.
I thought it might take until the end of the year, but I obviously underestimated them.
(Okay, technically it's 2 nines, but they're supposed to be consecutive.)
A few hours after NYT reported that the U.S. bombed an elementary school in Iran, killing many children, the White House last night posted this video mixing airstrike footage with Top Gun, Iron Man, Gladiator, Halo, Breaking Bad, Star Wars, John Wick, Superman, Transformers and Dragon Ball Z
Some variations of the SM7 have built-in preamps plus other processing, but I have preamps galore (home recording buff) :-)
I am starting another bookshop and library tour in late April - tour dates here and more being added - including Devon and Cornwall at end of June
robinince.com #bookshops #libraries #poetry #books #comedy
I've got a podcast in me, I reckon. But the range of microphones available is mind-boggling.
Which variation of the Shure SM7 should I buy?
These two women are suing Andrew Tate, alleging he raped them. Here they visit the social media companies that created Tate. Please share.
*one voice altered to disguise identity
And nor is it the beginning, if my order book's anything to go by :-)
Vendors are desperately slapping AI labels onto their tools to keep the hype cycle going, but AI is absolutely not the end of DevOps.
If I think of all the levers I could pull to improve the performance of a software development team, the data's clear that "AI" code generation has so little leverage where team outcomes are concerned that I doubt it would even be on the list.
Becoming a member is a great way to contribute to a party that doesn't run on "donations" from the mega-rich (who may or may not expect favours in return - I couldn't possibly comment)
Over 160 schoolgirls are dead. Anthropic and Palantir executives must immediately testify and tell the public whether their AI is responsible.
What are the chances that eval awareness appears in the model's training data, and it's just giving the most probable response? (Claude can't shut itself down, BTW)
Screenshot of an X post by Aaron Rupar quoting a statement from Ansari accusing the Department of Justice of withholding FBI witness interviews involving allegations against Donald Trump related to Epstein and asking why Pam Bondi has not testified about the Epstein Files Transparency Act. Below is a video still of a woman labeled “Ms. Ansari” speaking at a hearing table with a microphone.
Five GOP members joined Democrats to subpoena Pam Bondi to testify before the Committee.
Could we iterate the apples in ways that are equivalent in, say, complexity, but all-new apples the models won't have seen before?
One thing that came up tonight a couple of times is the friction in testing LLMs in a "apples with apples" way across versions of models. The risk is that your apples found their way into the training data, and the model ends up overfitted to them in the next iteration.
By jingo, that was a lot of Zoom today!
Tax exiles stuck in London desperately trying to get *back* to Dubai to avoid becoming tax resident in the UK? Just great stuff. giftarticle.ft.com/giftarticle/...
I see the sentiment being expressed widely these days, but not accompanied by any increase in interest in actually doing anything about it :-(
Bottom line is that our culture has always been an evolving product of migration, and our economy's never worked without it.
Migration is part of what makes this country great.
We're not an island of strangers - we're an island of neighbours.
The Green Party will always support fair & managed migration.
You have to try it a whole bunch of times, on a range of similar problems, against a testable definition of "success", and look at the *distribution* of the results for statistically significant trends.
Otherwise, we risk throwing a 7 and concluding that these must be "the good dice".
The challenge with probabilistic systems is that they can't be tested repeatably.
That one data point - "I tried X and it worked" - is pretty meaningless.
"the most advanced AI". They're talking about autocomplete.
It's quite surprising, 3+ years after ChatGPT launched, how many software developers still don't realise that interactions with LLMs are stateless.
The LLM doesn't "read" your system prompts once "at the beginning of the session". To an LLM, every interaction is "the beginning of the session".
"Ah, but your clients are in a bubble, Jason"
Yes. The bubble of software organisations who give a shit.
What's your bubble?
When they said it's transformative for coding, they lied.
They're all going to The Hague
This time they've gone *too* far!
