Posts by Tim 11 515 publicly visible posts • joined 4 Jan 2010
It's a failure of basic engineering practices.
Bugs like this could happen (and have happened) in many systems which allow users to search over both public and private data.
It can be *SOLVED* quite simply by ensuring that the AI model used on behalf of those outside the organization does not have access to private repos
In the late 90's I once had to fit a £3000 video capture card to my PC
I selected "shut down" from the windows start menu then proceeded to open up the case and poke around with the screwdriver to install the card
When I returned to the front of the PC I saw the message "it is now safe to turn off your computer". The PC had a pre-atx motherboard so the power had been on the whole time.
Fortunately everything worked OK
we're talking about something that's basically a CRUD data entry system with some document attachments, handling maybe a few million pieces of data, with thousands or maybe tens of thousands of requests per day. This is something that should cost tens of thousands to build in a few months, not hundreds of millions
Anyone who is deliberately ignoring AI (in their work or home life) is in serious danger of falling behind. If you ask tech people (especially developers) what they think of it, you'll usually find that the more they use it (and the better they learn how to harness it) the less cynical they have become. This is certainly true of me.
However there's a huge difference between using it as a tool and trusting it to make important changes and decisions, and that's where a lot of the AI evangelists are going wrong.
There are plenty of issues with AI, and plenty of areas where it's counterproductive at the moment, but the step-change in working practices is so distinct that anyone who decides wait until the wrinkles are ironed out before taking the plunge is in serious danger of getting left behind. This applies to individuals as well as every size of company.
For sure there will be dead ends but we can never go back to the old world now.
"spit out some variation of what it has been given to ingest" - that sounds very like a description of porting to a new environment
I'm no AI fanboy by any means, but COBOL is a pretty dumb and verbose language. Today's AIs should be pretty competent to port the code, including generating automated tests and verifying the new code passes.
Also don't forget "Sharepoint groove" which was renamed to "Sharepoint offline workspace" and then renamed to "OneDrive for business" (which obviously had nothing in common with OneDrive)
I don't think those other examples like twitter or meta are even in the same ballpark as Microsoft's constant switching of product names.
Rebranding a company happens all the time and is the preserve of the marketing department - everyone says it's stupid and moves on.
Renaming products so that two distinct product sets with different features are now called the same thing causes massive difficulty for users and IT staff, especially when you quietly withdraw one of them.
don't forget hotmail has also been renamed to outlook - although you can't access it via outlook.office.com; you have to access it using outlook.live.com (I mean, outlook.office.com will let you log in with your hotmail credentials but will then tell you that you don't have an account)
This exactly.
As someone who splits their time almost equally between AWS and Azure, It's impossible to overstate the difference in reliability between the two providers.
This also applies to accountability - it's difficult to beleive Microsoft would even be able to diagnose what went wrong, let alone whether they would be prepared to explain that to their customers. The attitude of "if anything goes wrong just try it again and it might work" applies throughout all MS software and services I've used.
As a freelance developer, what moved me back to Windows is that there are some customers with some apps that still aren't 100% portable (asp.net pre-core, PowerApps custom controls, and even excel spreadsheets).
Most of these can be overcome if you're persistent enough but I don't get paid for trying to fix my development environment so I need to spend my time being productive.
If you really need both Windows and unix-like, I think WSL gives the most seamless experience compared to dual-boot, parallels, wine, vmware etc.
IMHO there's 2 approached to management.
If the job is relatively unskilled or predictable you can have a command-and-control model where the boss is in charge and treats the subordinates as resources.
However if the term members are highly skilled knowledge workers (which should be the case in most IT teams) then the management should be more of an administrative function who's responsibility is to ensure the self-organising team has a clear understanding of the priorities and removes impediments to them working at their most efficient.
Most people are interpreting this content to imply that public sector IT is inferior to private sector, but I think that's missing the point.
Bad management and bad IT happens all the time in the private sector, but those companies go bust and the investors lose their money. That's the cross-check feedback loop.
In the public sector that can't happen so we really need to have better skills than in the private sector to stop those projects going bad
This strikes me as being quite similar to the situation with software warranties.
Software in general is so complex with such high probability of bugs that if we forced software manufacturers to warrant that their software was free of defects, they would just pack you and go home.
But such is the usefulness of software when it does work properly, that at the end of the day we all have to take the risk, and build safeguards into society to handle the inevitable failures.
If you believe "real" people contain some magic secret sauce undetectable by scientists then this may be true. If you believe the mind is synonymous with the brain then you have to accept that it's theoretically possible to build an AI that has the same mental capabilities as a person.
That's not too say that today's AIs are even close. As many have said, they work in a totally different way and are just machines for generating plausible-sounding sentences.
I have seen this in several apps and I believe the microservices mentality plays some part here.
It's trivial to set up a number of separate cloud data stores or APIs accessed through the same SPA, anonymously or with an API key, which encourages people to put all the business logic and permission checking in the front end (often with API keys hard-coded). I'm sure some naïve developers don't realise that even if your app has a login page, access to the underlying data store is not tied to the logged in user.
Designing a secure app is a lot more complicated than that and you'll have to either authenticate every data store and api using the user's credentials, or build a single authenticated API at the back end and funnel all your requests through that.
I've found that on ubran routes, organic maps (other OpenStreetMap apps are available) includes a lot of footpaths missing from Google maps.
The walking route feature is good at avoiding main roads and often taking you down footpaths and through parks, providing nicer surroundings at the cost of not necessarily taking the fastest route.
.... is that the way advertising will work in the AI world is the AI will adjust the answers it gives depending on who's paying for advertising, and it will be almost impossible to detect the degree to which this is happening.
There will not be a clean differentiation between the search results and the adverts like you get at the moment.
This is the real issue with consumer AI - the most effective way to get adverts out there in an AI scenario is not slapping advert boxes on the side of the search results; it's for the AI to skew its answers based on advertising. For all I know that isn't happening yet but it will do soon, and it will be a lot harder to work around than google's advertising model
What I find it is great for is if you want to do a simple task that you know lots of people have done before but you don't know the syntax or don't know the exact an API. For example "how to get all values from a set in javascript" or "how to generate a presigned URL to upload a file to s3 from python". these are things you could easily work out from the documentation but why bother?
once AI has generated the code and you've tested it, it's usally pretty self-evident whether it's correct.
However it is prone to hallucinations. It once had 10 goes at code to list the allowable keys from a typescript type at runtime before I decided to research and found it was impossible.
I had a really instructive copilot session yesterday when I asked it how to do something in typescript (iterate over the keys of a type). Over the course of our consersation it generated about 10 variations of a function, always taking into account my feedback and trying to fix it. None of them worked at all. I looked on stack overflow and discovered it was not possible to do what I wanted. I didn't have the heart to tell it.
the problem with the 'diversity' approach is that if you want a redundant system you usually want it to be identical to the master system in every respect so you have a high degree of confidence that it will work once failed over. If there are significant differences you'd need to re-test the failover every time there was a software or configuration change to any part of the system
I had a problem where Windows had hung so I couldn't use the restart option on the start menu
Every time I pressed the power button it just did a fast start and resumed into the broken state
even holiding down the power button for 4 seconds didn't work because the system had successfully hibernated before the 4 seconds was up
the laptop battery wasn't removable so I couldn't pull the power
IIRC I ended going into the bios setup screen and leaving it there until the battery died. then I could finally get a cold start
LLMs can be used in critical situations but only where it's possible for a human to evaluate the output.
Several times I have used LLMs to generate a small snippet of code where it was easier than wading through a load of StackOverflow posts or API documentation (e.g. date formatting in java!!!)
I'm sure Fujitsu has its problems like any large software development organisation, and maybe they could have done things differently in the Horizon case, but the scandal was down to outright corruption in the Post Office, and a law that let them get away with it, not down to anything within Fujitsu.
SQL server is probably the one piece of MS software I have very few criticisms of, and I think they have done an excellent job in backwards compatibility. Even when they made massive changes to the engine in 7 and 2000, they still provided an easy upgrade path
compare this to windows where they regularly delete important features with no regard for users' needs
In my company we used to get big customers and prospects call us at end-of-year to proactively look for discounted deals. Our response - "actually our numbers for this year are a bit ahead of where we'd like them so if it's OK with you we'd rather slip this deal back a couple of weeks so it goes into the next financial year".
This is all very true but I'm concerned by the fact that nowadays we seem to expect other people (even governments and large corporations) to take responsibility for our mental health.
Looking after your own mental health is not only possible; it's primarily your own responsibility and it's probably your biggest responsibility.
In the future I can quite imagine an AI would be able to examine the code to identify test conditions and generate a complete set of regression tests.
It couldn't prove the existing code was doing what the user wanted but could ensure nothing got accidentally broken by a bug fix or an update to third-party components or execution environment, which is the main reason we use test automation.
This could apply to end-to-end testing as well as unit tests.
