So what so “chat” is you are referring to chatting with, i.e., refining promotes, with a bot, not interacting with a human? Yes?
As for the video time consuming, yes, that can be true, but what is the alternative? The bot does understand the creative insight that comes out of the back-and-forth interaction that happens with humans. We are investing and creating in real time. The bot is pre-trained with patterns, and yes, it may be able to logically leaps, but it is not clear to me how the bot will be able to create attributes, write action code, and produce a customized template for the context you’re working on. Also, the video provides text/audio and visual sight that I think the bot might miss.
Talking of “The Tinderbox Way” - what happened to the rumours about a pdf update taking care of the many new functions that have been added since publication of the third edition? Did I miss something, @eastgate?
Amen! 18 years of documenting Tinderbox has taught me it never stands still. Even better, as artisanal software adding things that are useful to its users rather than bowing to the zeitgeist of the 99¢ AppStore ZOMG-so-kool apps.
Long story short. Written language != spoken language. I write with expertise of working on transcriptions of several years of an online (video) forums recordings. Transcriptions help understand the video, not the subject if used/read shorn of that concept. This is something the ‘AI’ boosters just don’t get. Getting a free answer we’d not pay for to a question is seductive. It’s sadly is not useful. For instance, much spoken word, verbatim, is composed of mostly incomplete sentences. Filling in the gaps often adds little meaning as the real meaning is inferred on-the-fly by the human brain. The AI brain fails badly ay the same task: we tried lots of offering and all same in a similar manner: software doesn’t understand human writing, and speech even less so. It’s a probabilistic attach on un0described semantics, that latter often depending on a context not on the page or in the spoken word.
Such software will get better, but outside narrow context, it’s not yet useful.
TL:DR; Will transcription help? No, unless you’re watching the video (eg. not knowing the speakers’s first language), but otherwise—not really. Enthusiasm alone doesn’t yield progress.
I’m tried serval different transcriptions, and frankly they’re not that great. And, separating the audio/text from the visual is often NOT as helpful as you may think.
I’m up for anything; however, do you have suggested services to look at?
This one below is done by Whisper.AI but locally using MacWhisper , I couldn’t figure how to turn on speaker diarization on with below. I feel honestly this is equal to YouTube auto transcribe
I think videos and other visual aids are helpful once you already have a basic understanding of something. However, as a new user, it’s important to start with easy-to-understand information. For example, I’ve heard that most people only use about 10% of the features in programs like Excel or Word.
My main goal is to do a basic job. You all have been using Tinderbox for a long time, so you create tutorials to help yourselves learn faster. The user is often an afterthought.
Using chat allows me to focus on what I want to know. If I just need to learn a keyboard shortcut, I don’t have to search for a video. Plus, video content isn’t searchable, so I don’t know what topics have been covered. It’s too much mental effort.
In chat, everything is explained in a clear and easy-to-understand way. Every teacher has their own style, and the student has to keep adapting to it. When the brain can’t generalize a specialized topic, it becomes overwhelmed and gives up.
Ideas and theories can certainly be put into writing. However, since these instructions come with visuals, I’m not sure how well it would work. What do you think?
Ouch. I think that is a little unfair. I think it’s actually to opposite. Faced with a multi-purpose app, as users we falsely assume our perspective/task is the same as everyone else’s and that it is ‘obvious’.
Indeed, but it turns out—inconveniently—that person A’s job and person B’s job are both basic but rather different. Years of helping people in this forum and its predecessor show that discomforting fact. Of course A may regard B’s task as irrelevant, and vice versa, but which then is the basic job?
Likewise ‘easy-to-understand’ is very subjective. We claim otherwise because it makes us feel smart and others are cast as dumb. In truth it is more subtle. One of the most interesting aspects of the 18-odd years I’ve been writing and expanding aTbRef is the correspondence with those who use it. People open fire with “not easy” but in almost all cases the problem is not that at all but one of looking for terminology/perspectives mapping their own. Once people know that their perspective is not everyone’s perspective they stop looking for a hidden key and rapidly engage with what is there.
Going the other way, as an author of help articles, such interactions with my readers have resulted in all sorts of improvements to the resource to try and give such people a handhold at start. For instance, an index is a great bridge but code-created indexes are poor as they don’t understand language, only word occurrence. But a human index for aTbRef would be a big effort and involve training I don’t have. And yet, discussion of indexing led to adding a site map where—as testified by feedback given—for many a Cmd+F will get them to, or close to the article/information they are looking for and fills the gap of a formal index.
It’s very interesting to hear your generally positive experience with chat—albeit involving the effort of curating your own LLM and asking for (free) content for it. I’m not sure about small tools/companies but your experience does suggest we might yet get helpful Help for MS Office or Adobe tools that we may interact with via chat.
I think an emerging area for study is how to write for both human and ‘AI’ understanding as they consume text differently and that fact is not without consequence if we pretend otherwise. I do wonder (is there any research on this?) how well LLMs cope with understanding richly interlinked hypertext where contingent reference is supplied by links. This is essentially aTbRef’s model explicitly labelled anchor text—the wiki model—makes for indigestible text as the expressive labels bend the flow of the descriptive narrative.
I don’t think any of this detracts from the value you seem to be getting from your local LLM and it is interesting to hear the outcomes, as it is still a novel approach.
My basic job is speed - I have like 30 mins on Tinderbox and if in that session, i am stuck looking for answers to use the tool rather than using the tool. The tool is coming in my way. I want this “speed/velocty” in every tool I look for.
The best way to engage with Tinderbox. Think: “what am I trying to do?” rather than “what’s the answer?”. Beyond the trivia of locating a menu item, most Tinderbox problems are more easily understood by the former not the latter.
Chat/AI is all very well but it pre-supposes the answer is known and has been written down and placed in the past-looking LLM. Most problems raised here are problems because the person can imagine what they want the outcome to look like but are having difficulty in describing it in terms of available features.
Also many things that are a ‘no’ as asked are a ‘yes’ if the problem is effectively described.
So ‘time saved’ is a false metric if you really want to get a task done, not least because it’s phrased in terms of others wasting our time rather than us wasting our own time by not pausing to understand our own question.
This doesn’t argue against chat and AI, but one thing is becoming apparent our faith in ‘AI’ ability to reason outstrips its ability to do so; garbage in, garbage out. An interesting emergent task is learning to write (input text) that an AI can more meaningfully understand, but on the human compatible side of writing in data triples.
Without doubt, I’m all about getting my work done as efficiently and effectively as possible. However, I often reflect on a U.S. Military saying, “Slow is fast and fast is slow.”
It most certainly took me a while to come up to speed on Tinderbox, but this is NOT Tinderbox’s fault. It is mine. I needed to learn abstraction thinking, metacognition, how to follow the data and ideas of data transformations, and several new languages and frameworks (RegEx, CSS, HTML), and most importantly, to reduce my residence, to stop looking for the "easy–do my work for me–button. As I evolve through this process, I’ve found that I can manage more much complex tasks, faster and more efficiently than I’ve ever done before. Moreover, the learning bleeds over to everything, not just Tinderbox. So, what I’ve learned is Tinderbox lets me make the choices, not give me the option to choose. In other words, with great power comes great responsibility.
Lastly…IMO “speed” is not a task but rather a method or process. I’m curious, what kind of tasks/jobs are you trying to complete?
Reading your post, three questions and an association come in my mind, in the wake of @satikusala :
What is your own ILModel (Idiosyncrasic Language Model — I am not sure this acronym exists —,it means the way you use to think and express yourself in your training activities, your relationship to knowledge too)?
Do you know it and did you try to understand and feel this aspect of your training pratice?
Why do not take time to explore slowly Tinderbox ?
A martial association: if you want to understand the way you have to do an oi zuki — an advancing basic karate punch — you have first to do it very slowly, as if you were learning taï-chi, refining the same movement step by step, day after day. Ease is not easy.
