My dog often gets misidentified as a restricted breed. This used to make apartment hunting difficult because, occasionally, the property manager would visually ID the dog breed as banned, I’d have to go to the vet and get paperwork, potentially gene testing, arguing she wasn't, it was a whole thing.
But, recently, the apartment I moved into had an online portal where I had to upload a photo and it would identify the breed to determine if it was approved.
I correctly assumed the portal was using an LLM for this purpose. I wrote a script which submitted different photos of my dog to the major LLM providers until it found a photo which all the LLMs would identify as the correct breed.
I simply submitted that photo and, as expected, passed with flying colors.
Interest has risen in tandem with the popularity of LLMs. I have worked on these sorts of things as part of internal tooling.
Sometimes these strategies are used to provide LLMs with a greater understanding of your codebase to improve code generation results.
Additionally, techniques which allow humans to visualize the shape of code are being explored as engineers become less familiar with the specific implementation.
“Before the Freedom of Information Act, I used to say at meetings, 'The illegal we do immediately; the unconstitutional takes a little longer.' But since the Freedom of Information Act, I'm afraid to say things like that.”
In a similar vein, in the pre-LLM era I gave a few interviews where the candidate was asked to screenshare while solving the problem. The candidates were allowed to use any resource they wanted from the broader internet.
I often found that I learned more about the candidate by the way they phrased their Google searches and how they selected and explored sites for information than from the actual solution they produced.
Friends and family who are largely non-technical have referred to ChatGPT as Chat for at least a year.
I admit I was struck when I first heard someone use that shorthand in conversation at a party. It was the moment I knew that for better or worse LLMs use had permeated deep into regular life.
I watched someone ask Claude to replace all occurrences of a string instead of using a deterministic operation like “Find and Replace” available in the very same VSCode window they prompted Claude from.
The US bugged the “toilet partitions” of the Russian embassy in D.C. during its construction[0] and the FBI built a tunnel under it for espionage purposes in the 80s[1].
I think it’s fair to say that diplomats appear to be appointed under a two-faced system.
On the one side you have some diplomats who really are quite capable career foreign policy wonks, appointed in a manner which appears to be meritocratic.
On the other side you have folks appointed, like you mention, as a kind of patronage.
Traditionally, it has been that the softer counterparties (Friendly countries, European allies, small island nations, etc) are staffed with patrons while the more difficult or geopolitically sensitive relationships are manned by professionals, but this is certainly not always true, and one can find many counterexamples.
This is interesting, do all major airports have the ability to set up this cable system if needed? Or is this unique to PDX, or perhaps only those airports which are near where fighters train.
This reads like a fluff piece for Goldman Sachs and the startup Cognition. It’s running in a bunch of outlets concurrently.
Goldman gets to appear as being on the cutting edge by incorporating AI. The startup behind the agent GS is using, Cognition (who is seeking a $2B valuation), gets to be seen as effective and bolster their name recognition.
The “AI Software Engineers” at my company, whose applications and workflows I often support as an SRE, by and large build things with LLMs rather than train or create them.
What they build ranges from product features to internal tools. They make heavy use of LLM vendor inference APIs, vector databases, etc. They end up writing a lot of glue code and software to manage the context of the LLM, query for data, integrate with other systems and so on. They also develop front-end interfaces for their applications.
Only recently have they started to, lightly, explore the idea of training LLMs with managed services like AWS SageMaker.
All this is to say that, the “AI Engineer/SWE” title will probably represent vastly different things depending on the technical sophistication of the organization.
If someone told me they were an AI Engineer at OpenAI I’d be more inclined to expect their role to be more fundamental, elsewhere, not so much.
He began the conversation by saying: “so as far as I know there’s two kinds of ai, open ai and closed ai…”
This speaks to your point, the people are confused.