Apple becomes the latest company to ban ChatGPT for internal use Apple has become the latest company to ban internal use of ChatGPT and similar products, ironically just as the OpenAI chatbot comes to iOS in the form of a mobile app. News of the move was revealed yesterday by The Wall Street Journal, which reviewed an internal Apple document informing employees of the ban. According to the …
Not sure what you mean. Unless it's FOSS, code is private IP. In some cases code leakage will hurt companies - and therefore their coders. It's not easy to separate what is confidential from what is not confidential, hence the outright usage bans on coding usage in those companies.
LLM systems that can be run in a closed environments will have a huge market. That means owning the context-specific LLM parameters (a separate layer from common general parameters) and managing them in the same way, with the same security, that the associated source code is managed.
I'm using Co-pilot in a context where it's perfectly alright to do so, for intentionally open software, and I rate it very highly as a tool. Very different from my personal experience using ChatGPT or Bard for internet queries and general conversation, for which LLM performance is still pretty wanting - generating fake references, etc.
(Flame!: What I really want from an LLM internet search is to get a list of references with minimal duplication, which I can further narrow to get closer to the topic I am targeting. But the apparent target for Bing and Bard is to offer a single answer, with fake or garbled references if asked. There are well known citation formats that every undergrad learns, with the understanding they cannot be fabricated - until an LLM "gets" these hard constraints, it cannot graduate).
... at <bank I'm currently working> for, the manglement has swallowed the AI hype, hook, line and sinker.
Not only is it NOT banned, its use is encouraged. Workshops held. Emails from all the way up to the CIO. Constant barrages about "how do we leverage this?" (while not shelling out for paid accounts, so we're stuck with the -3 models, instead of the supposedly far better -4)
Basically, having now had fairly significant use and exposure, my take away from that is it's "how do we drive the place into ground by using the hallucinations AI/LLM produces?" and "how do we best make our own jobs superfluous?"
It's mind-boggling!
Even just using it for the vaunted short-cuts to code, I have yet to experience a single instance, where I did not just end up rewriting from ground-up whatever CGPT produced, because it was not even wrong. I keep trying, I honestly do, searching for improvements, even just that one time where I state the problem and the code that comes out is actually copy-paste-able. Hasn't happened yet.
I don't think asking Chat GPT to write code given broad descriptions can succeed unless the task is really general. But most code is tied into context and not general.
Github Co-pilot is different - it's like a supercharged completion. It can often be pretty good at reading context and applying that.
Github co-pilot offer business subscriptions with advanced security - of course it costs but it's far better than getting ransomwared.
It's an interesting tool that seems to simplify the presentation of information. I have however found that links still need to be checked and information verified.
The results in a few cases has been misleading especially for some areas where there is limited information available on the internet. But that's to be expected.
The real issues arise when copying code into the engine as you lose control. Thus raising IPR and copyright issues as well as leaking business information.
There is also an issue with code snippets provided by the engines. They don't provide references so you have no idea what the copyright or licence maybe...
So in general it's a great tool for increasing the speed of research, but should be used with caution and don't copy code...
They need to resolve the "politician mode" issues*
"AI" will, especially when it has limited data on a topic or cannot interpret the question well, can confidently spout total BS.
Another nasty is lack of details - an answer can superficially seem fine, but may be missing crucial steps / information. Will seem fine to a "layperson" but anyone with a bit of expertise in the area can spot the (potentially catastrophic) missing information.
.. Though anyone who blindly follows an "AI" answer where the consequences could be bad** arguably deserves no sympathy if they make themselves a Darwin award candidate.
* Often more politely termed as confidently wrong / inaccurate.
** e.g. Probably not a good idea to ask "AI" about how to fix an electrical problem if you are not competent in that area (in UK, hardly worth doing your own electrical repairs as for anything non trivial need electrician to certify your work and they charge so much for the certification it's usually less hassle to pay them to do the whole job which usually is not often much more than the rip off certification cost (unless you can find someone to do the cert at "mates rates") and if sparky does it for you at least you save on the time it would have taken you)
Skill issue. naive thinking you can stop what is coming. the AI cat is out of the bag and its never going back in. become proficient and skilled at it or sit down
apple will become irrelevant if they cannot embrace the inevitable.
remember fear always arrives from what you don't understand. assuming you understand anything is entirely arrogant, so acknowledging that you're just pretending to not have fear for what you think you understand, should expel all fears. you never understand anything you just get used to it and start to take it for granted. So too will it be with AI.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
