Huh. Iâve not posted one of these since May 2024. Good job Iâve collected up a few to share out with my meaningless thoughts on each one. I think this roundup is quite electic. Lets get to it!
Technology
Some what meta comment: Is AI separate from technology, or is it all part of technology know? đ€ Anyway, Iâve lumped it all together.
Alignment faking in large language models: I thought this was interesting â not so much the concept of a model âconcealingâ itâs true thoughts, but the extracting some âsecretâ thought stack into a separate scratch pad. Itâs not clear from the post if this is in the token stream (and trained to skip over?) of if itâs distinct from the token stream, so doesnât influence the output token stream. On âconcealmentâ itself: It doesnât feel that odd given the large input data set that even in a âpureâ token prediction scenario that linguistically the concealment would be covered through text. It doesnât feel super insightful?
LLMs have say everything out loud. But maybe they donât: Iâve long been⊠displeased? Disappointed? That the core capabilities of the LLMs seems to be exposed through⊠english prose. Weâve come thousands of years, a 100 years of computers, with hundreds of formats for data and we conclude⊠english? Is the lingua franca of the future of intelligence? That seems weird, and terribly inefficient. But this article suggests that there is hope to being able to have the conversion to english/text be optional. This gives me hope, because surely the future isnât built upon a keen understanding of their / theyâre & your / youâreâŠ
Faith & Fate: Fuzzy Transformers: insert that standard distribution graph with the amateur at the start, enjoyer in the middle, and the expert at the end. Amateur & expert say âLLMs are just fancy auto completeâ. Enjoyers ascribe consciousnessâŠ
The AI Summer: Theres something different this time with AI. But itâs not that itâs going to be the fastest technology transformation â itâs going to take a long slumbering, lazy summer, to find out how they can actually be useful; how they can have purpose; how they can add value. How they can be something beyond a distraction.
Machines of Loving Grace: This is a great read as the contrast to the fever dream that is Situational Awareness. This seems to have its feet on the ground â albeit hopeful, aspirational feet â about what might happen. I think itâs too positive in its hope for biology/medical breakthroughs, and too hand wavy for the social/governmental impact (âidk, might be hardâ doesnât cut it). Definitely worth a read to at least level set how those that are running the foundation modelers are projecting their own dreams.
Where are we on AI expectations?: as they say âitâs artificial intelligence until it worksâ, this takes a look at some of the historical cases of the last few AI boom-to-bust and asks us to learn from the past
What Do We Want Computers to Do?: This is a great article that gets to the core of what the value of generative AI is. Everyoneâs own personal bar will be different. There are levels of care & passion that go into the various jobs we all do in our lives. But there is also context â if Iâm unskilled (in art, as I am), and I wish to obtain an image for a flyer etc, I have different choices available. I can pick from pre-existing stock art (from free clip art to a getty image), I can create something that does not meet what I envisioned, but is âauthenticâ, or I can accept something that gets closer, but still a long way from my minds eye (aka Generative AI). Itâs not an absolute measure â the context does matter. But itâs also influenced by people who have something else in their life that to them matters more than creating something without âgenerativeâ tools â either because of cost, time, or they dgaf. There is no one answer.
Software Engineering
Decision-Making pitfalls for technical leaders: Making decisions is more than just making the decision. Theres so much more about the context of the decision â not just the things that go into the decision, but your own framework of making that decision. Itâs very easy apply one perspective to all decisions, but really you have to think about what it is youâre actually trying to achieve and why.
Systems: The purposes of a system is what it does: Reading this article, I realised I do this all the time. I just wasnât thinking about it in a pure systems sense. I was caught up in âincentivesâ, and how when people rail against behaviour of their coworkers â a faceless process implemented by their coworkers â I say âthink about the incentives that the humans involved here have, and how they may cause the behaviours you seeâ. But this I think is so much better. The system is doing what it does. It is not âbrokenâ. The context, and purpose of the system might be broken, and instead of changing the system â wagging the dog â change what the system is supposed to be doing.
Poets and police: This feels like âDreamersâ vs âAbsolutistsâ, and Iâm so in the Poet/Dreamer bucket. I love the contrast between someone painting a picture of a better future â a path forward â and someone who calls out the gap between what is being sold, and the reality ultimately in an attempt to destroy the dream.
Software Engineer Titles Have (Almost) Lost All Their Meaning: There is a trope â one which I have observed first hand â that titles have been inflated over the last 5-8 years or so. I saw it before that with the phrase âSenior is the new SDE IIâ, and Iâve seen it recently as âPrincipal it the new Seniorâ. The reality is that itâs not that clear cut, but theres also more than a grain or nugget of truth here.
There is â of course â the counter-argument that this theory is powered by capitalistic motivations on company and personal levels.
- As an employer, you of course want to keep titles deflated â it saves you money!
- As a coworker, you want to keep getting the bigger slice of the pie, and to laud your title over others. You need to punch down to keep yourself elevated.
Eng org seniority-mix model.: Will Larson has been on a tear recently using Systems Modeling to look at the impact of different policy or investment choices you can make in your organization â theyâve been great. This one, however, shone a light on a slightly uncomfortable truth: You canât promote everyone all the time. You have to constrain and be intentional about your mix. You have to also manage those senior people out to promote from within.
This is an interesting companion article to âTitle Inflationâ Iâve also linked.
Writing Down (And Searching Through) Every UUID: I love how everything is âfakeâ. None of it is an anything like what you would build to meet the âspecâ because itâs all tweaked meet what your perceive correctly rather than the âfactualâ interpretation of the âspecâ.
For historical job experience reasons, Iâm still impressed he got the scrolling almost perfect, when there is, infact, no scrolling happening.
Microsevices are technical debt: Interesting video, spicy title. More seriously: âMicroservice all the thingsâ does lead to technical debt. But itâs better termed âsociotechnical debtâ rather than pure technical debt.
But it also sparked this thang Iâve been grinding on for years: Microservices are âdynamic librariesâ from client software development with higher latency. Itâs the same problems but with latency. For the longest time Iâve been trying to find the right way for me to talk about it like that without people going âwtfâ. This video actually made me realise: there is more active discussion of resolving these tensions in the âservices worldâ than the client world (E.g., big honking windows monolith), and that I should instead be trying to get the client monolith people to engage with the services world to make forward progress on understanding their problem.
(Post that took me to this video)
Leadership / Management / Career
Economic Termites Are Everywhere: This would have been better titled as âHidden Monopoliesâ. It opens really strong, and sets a great picture for an interesting deep dive. But then whiffs it by basically exposing some relatively obvious monopolies, rather than resolving the opening scenario of an expensive commercial space refit. Still an interesting skim.
Nobody knows whats going on: Working in tech, at a time that tech is written about everywhere, really throws this into sharp relief. Weâre all just moving around on hopium & copium highs.
Story Points are Pointless, Measure Queues: I have, for over a decade, used the phrase âstory points are not a convertible currencyâ to mitigate the inter-team battle of how-many-points-did-you-do that the Pointy Haired Types always catalyze. This article talks about this, and some other suggestions. Itâs conclusions are reasonable, grounded, and well thought through (breakdown to tasks, measure you queue length). But at the same time it feels like itâs written from the perspective of âThis is a pure way to do itâ, and misses how you can actually get it in to an organization. It also feels grounded in âmy problem can be mapped out without reams of pre-workâ, which I feel some âhigh-techâ companies canât do â bringing up a new AI model integration into an OS seems like itâs not a âdiscovered set of workâ, and makes it hard to extrapolate further out in reality.
Things I learned about strategy: While short on âdataâ, this post resonates. âYou make things, or you sell thingsâ is a truism â I always thought âdonât work in a cost centerâ was also a good way to apply this. And the strategy â so much yes. âstrategy is not something that some staff person does, strategy happens every day in every decision and your line managers need to be your best strategistsâ.
Grifters, believers, grinders, and coasters: The article says this, but before you click: This isnât about deception or fraud in the workplace. You need to go in thinking of different words to describe the archetypes. I would have picked âPleasers, Believers, Grinders, and Mindersâ â Pleasers want to present a perspective that ensures they move forward by pleasing people; minders are just minding their own business doing whats needed and getting on with life. Anyway, the article talks about the different relationships that they have with their jobs, and how you need a good mix for a healthy organization.
Why shouldnât I go back to my ex?: âWhy shouldnât I go back to my ex?â He said, âIf you see the same tree twice in a forest; Itâs because youâre lost.â. I canât stop thinking about this â not in the context of romantic relationships (I donât dispute in that context!), but life in general. That job you once had. That house you lived in. When youâre searching for a repeat, it means youâre not sure where you are any more and you need comfort and familiarity.
How AI will change democracy: The interesting take away isnât âAI & Democracyâ, itâs that automation has outsized impact when it enables order of magnitude changes (increases) the scale the system or process being automated. You can see echos of this across other times â where scale has changed through automation. Publishing (the internet) becoming free â changes everything about information flow. eCommerce changes how prices manifest â they change in a blink of an eye, and are even less uniformly distributed than âbeforeâ. Be conscious of what a step change would mean.
The Ghosts In The Machine: Musicians are taking contract work to make ends meet and feeling the loss of their âsoulâ is the same software developers doing their job where they have become unmoored from passions - side projects are their true passion. The musicians are aghast at the fact that business went and businessed something that was clearly going to happen, even without Spotify/Streaming. Boy/girl band formulaic pop of the 90s-00âs for told this. Bowie was the harbinger with Bowie Bonds. Weâre now surprised by the overtness, but it was inevitable.
Casual Viewing is essentially the same, but about the movie industry. Itâs more impassioned, more existential, dare I say ranty. But itâs come across the same point â commoditization of content, because it turns people donât care about what they read, what they watch, and what they listen to. Itâs not chewing gum for the eyes or ears â itâs chewing gum for the soul. And itâs full of empty calories.
Home-cooked software and Barefoot developers: Iâve long had this vibe that a lot of software is built solely in the context of âhitting it bigâ or being âa successâ (by the larger social norms), and that weâre missing the fact there is a different world for software that âsolves peoples problemsâ â problems that are terribly narrow, and absolutely not a business. But they do make peoples lives better. This presentation/transcript from Maggie goes down this path under the banner of âlocal-first softwareâ (really, she means local software for local people) with a side helping of a concept of âbarefoot developersâ.
Design doesnât have to end like this: A very thought provoking article about the relationship between business, AI, and the discipline of Design. Theres a thread of âstick it too the manâ, and âitâs about the craft manâ. But then I realized that if you substitute âdesignâ and âdesignerâ with âengineeringâ and âengineerâ, and it hit differently. It was clear I was saying the same things, taking the same stance â unwilling to embrace the inevitable, and thinking about âtasteâ as the de facto determinator in everything.
How I use âAIâ: This is such an interesting article. Itâs how a real person really uses AI/LLMs. And it blows my mind in a way that I have to introspect a lot. This person deals in a currency that feels somewhat ephemeral â nothing is permenant. And the LLM does a great job here. But also seeing the level of detail and discussion they go into throws me off a little, and I canât quantify why. We invest orders of magnitude more effort & time to get actual fucking humans to do this stuff. But this â chatting with an LLM â some how feels like more work for less valuable outcome. But, thats just the feeling.