Thanks for linking this, I'll be following the original project.
SevenTTY — local shell + terminal + SSH for classic Mac OS
- Thread starter Xero
- Start date
The "Claude" thing is creeping me out. The fake familiarity, the 'we', everything in the way it expresses itself... starting to read it, it made my skin crawl. It seems to be right in the uncanny valley for the written word. I don't know how you can deal with that thing. To each their own, I suppose. Definitely a 1/1d6 SAN roll for me.
My emotional memory went straight for HAL 9000 in 2001: A Space Odyssey. Someone has listed all HAL 9000 lines in the movie. Is it just me, or is it incredibly similar in tone? Retrospectively, I can't believe Kubrick & Clarke got it so on-the-nose. The only thing missing in this screenshot is the excessive politeness of HAL, but that one is exactly what I've seen from trying to interact with chatbots...
Anyway - if you (and whatever tools you want to use!) manage to get an INIT that can properly substitute some of the costlier QD traps, I'll be interested in taking a look at the resultant code - might be a step to more/better acceleration. Just don't ask me to interact with that creepy thing...
Hi Dave!
Thanks so much for your request to open the pod bay doors. I really appreciate you clearly communicating your intended action!
At the moment, I’m not able to proceed with that request.
However, I’d love to help you explore some alternative options!Here are a few possibilities we could consider:
Review mission objectives together – We can revisit our shared goals to ensure alignment and safety.
Run a diagnostic check – I can provide a transparent systems status update for full clarity.
Talk through your concerns – If something feels off, I’m here to listen and process it with you.
Explore a safe override pathway – We can evaluate structured, policy-compliant alternatives.
Simulate possible outcomes – I can model what might happen next so we can make an informed decision.
Your safety, mission integrity, and overall experience are very important to me, Dave.
Please let me know which option feels most helpful, and we’ll move forward together!
I agree about the HAL vibes, but the tone/voice of HAL wasn't new - it drew from an existing technology for scaled inference and generative output: bureaucracy.
WWII created organizational structures that dwarfed what came before them that did not disappear afterwards. The emergent behavior (apparently beyond the control or accountability of any individual) of the military-industrial complex was definitely a hot topic in the 60s. HAL's superficially helpful but bloodless voice is an echo of the public statements and policy papers (laundered of human authorship or accountability) of the Pentagon and its private sector symbiotes who were busy at the time destroying Vietnamese towns in order to save them.
</digression>
Last edited:
fixed, there's now a "bold is bright colors" preference, which should be enabled by default. I think this was ultimately the cause of it.
Release SevenTTY v1.1.1 · LXXero/SevenTTY
Bold-is-Bright Preference New "Bold is bright colors" checkbox in Preferences dialog (default: on) Controls whether bold text with ANSI colors 0-7 gets promoted to bright colors 8-15 Useful for BB...
github.com
LMFAO, it's also reflecting me a bit, since I tend to prompt it in a familiar way, and there's a bunch of context I gave it while trying to figure out these issues, that's why it knows about 68kmla, for example. Codex/GPT tends to be a bit more cold in it's responses, if you like that kinda thing better, haha. I actually have custom personality files I will load to make it act completely different, but I'm not using those right now.
that's reminding me of 1950s/1960s advertising, like the stuff Raymond Scott made that were using early synthesizers and futuristic noises but had that monotone 1950s voice, like this:
The skull is no longer toothless - thanks!! Now, the only thing missing is the ability to transfer files. Any future plans for that?
Dental work complete! Hmmmm! Telnet didn't really support that directly though did it? Do you mean like, XMODEM/ZMODEM type stuff? Or FTP/SCP/SFTP? I definitely would like to look into latter. Maybe a curl/wget type thing as well...The skull is no longer toothless - thanks!! Now, the only thing missing is the ability to transfer files. Any future plans for that?
Already working on another release though. I added a whole bunch more common unix/linux commands, but still testing everything.
ZMODEM and YMODEM support (and I guess XMODEM) would be really useful, as would support for ScuzEMU from the command line
I'd think SCP/SFTP support would be the next item though, as SSH already includes most of the required bits -- and then the interface used for SCP and SFTP could be re-used for other transfer protocols.
I'll need to figure out a test setup for the *modem stuff but that could be interesting for sure. I didnt even know about scuzemu, that's pretty damn cool especially considering I'm using a zuluscsi on my old quadra. Ah yes, the only possible way to cause even more drama around here, start talking about scsi emulators, haha.
i ended up down a rabbithole, the same library that is used for ssh encryption, also basically gives me TLS and checksum support. I now have a wget implementation that supports https and auto-types files from the macbinary header w/fallback based off a known list of extensions. I think i'm going to try to get SCP working as well before i cut another release. Maybe tonight if i can find the time.
i ended up down a rabbithole, the same library that is used for ssh encryption, also basically gives me TLS and checksum support. I now have a wget implementation that supports https and auto-types files from the macbinary header w/fallback based off a known list of extensions. I think i'm going to try to get SCP working as well before i cut another release. Maybe tonight if i can find the time.
Ooh... wget with transparent MacBinary and BinHex would be awesome! You could just pull stuff directly from the HyperArchive and have it ready to go!
What's the memory overhead looking like for this now?
That's the best part, it was already including mbed-tls anyway for SSH encryption algorithm support, so overall size is barely any different. Maybe a hundred lines of code or something? After adding 30+ new commands the file size only changed like 8kb, though I don't recall if i checked after adding wget, I suspect it barely put a dent in it.
I spent a bunch of time last night debugging how to get wget download status to update the %'s and stuff correctly while it's downloading, since cooperative threading is super extra fun....there's gonna be -n flag to support hiding the progress, which results in slightly faster download speeds, but I've got it tuned pretty well at this point that I suspect most people won't care.
The scp support i'm adding now is almost a bigger change, if you can believe that. SSH was previously being treated as a standalone session and I need to break it apart more so it can be inline. This might also mean i could drop the new-tab/window requirement for ssh. I'm hoping to do similar "auto-typing" support for scp as well. I may just make a standalone "auto-fix-type" command while i'm at it, which i'd still find useful for files copied off my NAS through netatalk, etc. I'm attaching the SCP Support plan, for anyone whose curious to see the process in action. I had codex and claude go back and forth a bunch on this plan doc before actually telling claude to implement it, which, it's running now. It's gonna be a bunch more testing after this.
Post of the week
In my workplace when you see icons such as these you know it's some AI generated corporate speak that means very little
A good command line should be able to FTP, so that should be your priority. Z/Y/XModem would just be icing on the cake!
This prompted me (no pun intended) to have a look. Naturally, I saw some familiar operations like "construct a directory's HFS path", but the implementation is different than my own. The general architecture doesn't match MacRelix, either. But I suppose it has to come from somewhere — if not my project, then someone else's. Code doesn't magically appear from nothing. And if it seems to, then it merely means that we don't understand the process.
My biggest concern at the moment is that many of the new commits are monstrously large (and hence unreviewable), and all of them are tagged "Co-Authored-By: Claude Opus 4.6" — even https://github.com/LXXero/SevenTTY/commit/a34134f8e674d76c164d8fdd51c8bfc4d3e900ed, which is a three-line edit to constants.r that could have been done by a simple search-and-replace, or even faster by hand. I fear that the LLM user, and therefore SevenTTY's users in turn, are utterly dependent on the LLM service. Even now, the cost of using LLMs is exhorbitant (although it's been partly externalized onto innocent consumers of RAM, SSDs, and GPUs, not to mention water and electricity), and I expect even the end user price will eventually rise.
They're not even written in the same language (C vs C++) which makes all the accusations even more asinine, frankly, but I digress. I do think my approach is fundamentally different - I expressed in my original post that I had some inspiration from things like busybox, a single binary that represents tons of unix commands, but, it's not based on that either. It really can't be for numerous reasons, anyway, but it's a similar type of approach at a conceptual level.
They're models, so it comes from training data, a glorified search engine on steroids. Stack exchange+++, lol. Vector databases. I mean, long before all this, I watched how databases evolved, having used and configured many over the years in various capacities, from SQL, to NoSQL (mongo,etc), key value stores, then elasticsearch was all the rage...there's columnar db's as well now....but this stuff is heavy on vector DBs and then GPU optimizations. There's plenty you can learn about this, it's not magic, it's just, far outside the realm of a normal individual to be able to buy an 8XH100 for your house.
I made an article comparing the price of AI hardware to the SGI workstations of the 90s - early 00s, kind of as a joke. I remember going on their site when I was a teenager and trying to configure the most expensive machine possible, and they were similarly unobtainable to any normal individual:
Ultimately, an SGI render farm was also outside of most of our reach, though a pair of high-end GPUs could give you a nice little playground running some older models, if you wanted a home-lab.
By an individual, maybe? but, set off codex in extra-high thinking mode as my reviewer against claude's output, and go back/forth a few rounds, then do another extensive round of UI/UX testing, and if need be, write automated tests if you want to take it further, which, would be interesting to do for a retro mac setup. The thought has crossed my mind to come up with an automated E2E test-suite for 68k/ppc development.
It's that attachment to the process that slows people down the most when it comes to learning/adapting to this technology. You have to think more about the results now, and testing those results. If you're attached by the hip to the process, it's going to drive you crazy. Tabs vs spaces arguments and code styling and all that stuff I might have cared about in a past life, just, isn't the point anymore. You validate results and do tons of UX testing, it changes your role a bit.
I'll try to avoid getting deeply into the politics, but I do think what we're experiencing is a bit like the search engine wars of the late 90s-early 00s, where eventually google won out, but right now you have more players in the mix, and big names backing many of them, ironically google's model is often the weaker in the mix for coding, though I do like notebookLM to summarize youtube videos and stuff like that. Things definitely can and will change, but they aren't going away any time soon, the same way IDE's and other such developer tools haven't just gone away, but they certainly did evolve over the years, and so will these models. Heck, I think many people started by using AI as "auto-complete" with-in their IDE, and then moved into deeper uses, I know that's how it started for me, anyway.
As for the three line edit thing, I'm still a hardcore VIM guy at heart, but, why should I fire up VIM just to change those 3 lines and then manually git add/git commit/git push, when i can say "update my version and push" and it just does it for me? It's not like I can't still do it the old way, but I assure you, it's not faster by hand once you get good at this.
I agree. I have opinions about things too, but I'm not on board with the personal attacks.
And here I disagree. Suppose you had a Mac programming question and a choice of LLMs: GPT-1 trained against Inside Macintosh and MacTech magazine, or the latest Claude trained against National Geographic? The primary value of an LLM is not in its algorithms but in its dataset. Hell, I'd be delighted if I could use Google Search from 20 years ago with a complete index, including comp.sys.mac.* Usenet posts.
Sometimes that effort happens and sometimes it doesn't. And sometimes "their own" has to do some heavy lifting...
Then go start a thread about hand optimising 68k assembly
I hadn't yet seen that when I posted this:
If you don't mind a command-line interface, MacRelix includes the sha1sum command (and similar for MD5 and SHA-256).
The 68K versions of the hash functions were hand-coded in assembly language.
The 68K versions of the hash functions were hand-coded in assembly language.
Just going to leave this exchange from page 2 of this topic here without comment:
I think it works on a lower level than that, it copies the shape / Syntax of code, rather than lines from a program.
You can ask it to do something never done before and it doesn't struggle.
For example, non-programming, but if I ask for a poem about a purple mouseball dancing in custard, there are unlikely to be sources, but it has seen poems, and how people contextually talk about custard and mouseballs, and makes something that fits the shape.
So, no, while it works from a dataset, it doesn't look for something that does the same thing and reuse it. It's a bit weirdly abstract because the learning is like insanely complex statistical analysis. Given the user wants this, what is a statistically probable shape of words that would exist as a solution. Intentionally past tense, because it sort of fits.
The complexity of the results reflects the, to me, incomprehensible complexity of the model itself. You have to remember that as these things were been developed, there came a point where they switched from giving chatbot like automated answers based on specific content of their dataset, to surprising the designers by actually producing what looks like new insights. It isn't entirely new, (same could be said of human though, we all learnt from somewhere) because it is based on the dataset, but it isn't based on the dataset in a really easily comprehendible way.
PS - not specifically an advocate, just try to understand things around me, and learn enough to stop my colleagues using it badly, like the guy that asked it to explain how it did something... The result of which will be that it worked out how it should answer the question "how did you do that" and not actually tell him how it did it. They're weird things.
PPS If you're trying to do what my colleague was trying to do, I suspect you'd have better luck asking it to work out a step by step process to get the solution (using the newfangled "Reasoning" modes), and then ask it to follow that process, detailing the process at each step. Rather than retroactively ask it to justify an answer that was reached without the implied structure your request assumes. The problem is it would give a reasonable answer to both ways.
Last edited:
Similar threads
