AI: The Rise of the Machines... Or Just a Lot of Overhyped Chatbots?

pwe

Silver Baronet of the Realm
1,303
6,894
I am completely blown away by Claude Cowork. From this simple prompt it finds the 'Pets' thread and gives a nice pet description, including context. It uses a background browser and does actual mouse clicks to navigate and take screenshots. So far it successfully did everything I've thrown at it. I gave it my LinkedIn profile and asked it to find relevant jobs and write PDF applications. It did it perfectly and created some excellent applications.

In the last section it says Mahes also posted a photo of his mom (who was also quite fluffy). I am assuming it refers to the cats mom.

1779264081695.png


1779264166748.png


1779264196279.png
 
  • 1Like
  • 1Worf
Reactions: 1 users

rhinohelix

Dental Dammer
<Gold Donor>
3,570
6,040
Redundancy between data centers is a backup, very literally... did you mean to phrase this another way?
In ITIL terms, no one ever refers to a live copy of data on an active server, whether its on a passive server in an Active/passive pair, or a HA set of servers in another DC where the data is being synced or changed as a "backup". A backup is something physically or logically separated from production changes that can be restored in full or part if need be. I mean, it goes deeper than that but just having another set of servers somewhere doesn't count, and depending on the way they are being synced might also be affected.
 
  • 2Like
  • 1Solidarity
Reactions: 2 users

Khane

Got something right about marriage
21,900
15,872
In ITIL terms, no one ever refers to a live copy of data on an active server, whether its on a passive server in an Active/passive pair, or a HA set of servers in another DC where the data is being synced or changed as a "backup". A backup is something physically or logically separated from production changes that can be restored in full or part if need be. I mean, it goes deeper than that but just having another set of servers somewhere doesn't count, and depending on the way they are being synced might also be affected.

How many times can one contradict themselves in one paragraph? This isn't even really your fault. I blame the nebulous, dubious language that has steered technology "paradigms" over the last 30 years.

"It's not wifi! It's mesh! It's totally different!"
 
  • 1Picard
Reactions: 1 user

ToeMissile

Pronouns: zie/zhem/zer
<Gold Donor>
3,769
2,549
Meanwhile, since ~2004 you could just set a reminder yourself on your cell phone in about 10 seconds total.

The things people waste electricity and compute on with AI is a thing of beauty.
This is true, and I get it. I tend to be forgetful and dislike managing the minutiae of life. It’s as much just fucking around to see what I can get working and what will actually be helpful so I can spend more time with family/friends/etc.

weird stuff like the screenshot along the way is just entertaining.
 

Khane

Got something right about marriage
21,900
15,872
This is true, and I get it. I tend to be forgetful and dislike managing the minutiae of life. It’s as much just fucking around to see what I can get working and what will actually be helpful so I can spend more time with family/friends/etc.

weird stuff like the screenshot along the way is just entertaining.

I guess the point I'm trying to make is people use AI in the wrong way to try and simplify their lives. It's like AI makes them forget how anything and everything works.

So using that example, prompting AI to do this type of thing can be efficient and a time saver. However people prompt and direct the AI in exactly the manner your screenshot depicted instead of how they SHOULD be prompting it.

Which is:

Add this item/event/occurence to my calendar app on this date and at this time. Add this note/description to the event. Set a reminder for it 1 week prior, 1 day prior, and 1 hour prior.

It really is that simple and wastes far less time, compute and actual mental AND physical, tangible energy (electricity) by just giving it actual, discernible tasks. And it fucks it up way less.

Telling it to just remind you will almost always result in it trying to create some funky script with its own reminder logic that may or may not fire, versus just having it add a task to something that is a known entity, works and does exactly what you want it to do.
 

rhinohelix

Dental Dammer
<Gold Donor>
3,570
6,040
How many times can one contradict themselves in one paragraph? This isn't even really your fault. I blame the nebulous, dubious language that has steered technology "paradigms" over the last 30 years.

"It's not wifi! It's mesh! It's totally different!"
Dude.
There is literally nothing contradictory in what I wrote. If you don't understand what a backup is or is supposed to be, no worries. Not colloquially, like "I got a spare" but actual recoverable data from a fixed point in the past. Production data, or data that can be changed in a live environment, isn't a backup. A backup is something physically/logically distinct from production data. Even a standby database, if its being synced with live data to be used for recovery, isn't a backup. That can be Redundancy or Recovery or High Availability, and all those terms mean specific, different things but none of them are backups. They have consequential meaning in enterprise IT, like getting your business back up and running and losing millions of dollars or not.

I am not as involved in the day-to-day management as I was in when I was in DC/Service Delivery but we used to back up critical systems to physical media, keep x number of days onsite, and then rotate some offsite to storage at Iron Mountain. I know Cloud storage is much more ubiquitous now but my career moved in a different direction so I don't spend as much time on that part of landscapes these days. 3-2-1 was and is the Iron rule of backups.

If you google "ITIL definition of a backup" you mostly find folks working from the ITIL v3 framework. They are on v5 now but things like that never change.
 
  • 1Truth!
Reactions: 1 user

Khane

Got something right about marriage
21,900
15,872
I want you to read your post, the one that I quoted, from the standpoint of "words have meaning". And then contemplate how contradictory the entire thing is.

Its magical how brainwashed people become in the tech space with buzzwords.

The best possible example of this is virtualization, what it was designed for, and how it has never, ever been leveraged for its intended purpose. And how Cloud service providers are robbing everyone by tricking people into thinking the way you do.
 
  • 1Picard
Reactions: 1 user

rhinohelix

Dental Dammer
<Gold Donor>
3,570
6,040
I guess the point I'm trying to make is people use AI in the wrong way to try and simplify their lives. It's like AI makes them forget how anything and everything works.

So using that example, prompting AI to do this type of thing can be efficient and a time saver. However people prompt and direct the AI in exactly the manner your screenshot depicted instead of how they SHOULD be prompting it.

Which is:

Add this item/event/occurence to my calendar app on this date and at this time. Add this note/description to the event. Set a reminder for it 1 week prior, 1 day prior, and 1 hour prior.

It really is that simple and wastes far less time, compute and actual mental AND physical, tangible energy (electricity) by just giving it actual, discernible tasks. And it fucks it up way less.

Telling it to just remind you will almost always result in it trying to create some funky script with its own reminder logic that may or may not fire, versus just having it add a task to something that is a known entity, works and does exactly what you want it to do.
You are not wrong; Prompt engineering is really not as off the cuff as lots of people think it is/as it seems. What passes many people by is that it's just another abstraction layer from coding. In the distant past, we used to write directly in machine language/assembly. Then we came up with C/Fortran to move one layer up and give instructions to chips to tell them what we wanted to be done. Now we are yet again moving back another abstraction layer crafting prompts to have LLMs assemble the code to tell chips what to do in their own language.

It's a fair amount of crafting to get what you want, and your example is a great way to put guard rails on the simple request. As it gets more complicated, you have to add explicit language for guidelines, give the LLM a role, tell them tone and limits, etc. The more detail you add, the more you can get out of the prompt. There is, of course, an inevitable limit, along with all the noted pitfalls and dangers.
 
Last edited:
  • 1Truth!
Reactions: 1 user

rhinohelix

Dental Dammer
<Gold Donor>
3,570
6,040
I want you to read your post, the one that I quoted, from the standpoint of "words have meaning". And then contemplate how contradictory the entire thing is.

Its magical how brainwashed people become in the tech space with buzzwords.

The best possible example of this is virtualization, what it was designed for, and how it has never, ever been leveraged for its intended purpose. And how Cloud service providers are robbing everyone by tricking people into thinking the way you do.
What is this magic intended purpose?

I have been working with virtualized servers for more than two decades now: How have we been led astray?

Edit: That's just the terminology of the industry. Doctors have terminology. Geologists have terminology. Architects have terminology. IT, specifically Enterprise IT, uses ITIL terminology so that we don't have issues like this and conversations don't become a mishmash of "Who's on First" . It's literally why the ITIL framework was invented. It's not buzzwords or marketing speak; it's how industry professionals communicate. We fight constantly to keep our company (whichever company that is, its a struggle at every company at which I have worked , and I have worked for a number of the biggest) to keep them aligned and not doing their own thing.
 
Last edited:
  • 1Truth!
Reactions: 1 user

Khane

Got something right about marriage
21,900
15,872
What is this magic intended purpose?

I have been working with virtualized servers for more than two decades now: How have we been lead astray?

There is no magic. The purpose of virtualization was to decouple software from hardware, thus making the software durable and easy to migrate. Physical vs logical as you kind of touched on. Yet still to this day we have companies deploying entire code bases to multiple regions AND disaster recovery sites in active, active, passive setups AND still requiring antiquated tape backups AND now with the added bonus of paying exorbitant premiums for "virtualization" of cloud based hardware and services that always have to be on so they can sync but somehow dont count as backups. And new white papers and nomenclature and buzz words furthering the grift every year.

Maybe I was being coy, or perhaps obtuse?
 
  • 1Picard
Reactions: 1 user

rhinohelix

Dental Dammer
<Gold Donor>
3,570
6,040
There is no magic. The purpose of virtualization was to decouple software from hardware, thus making the software durable and easy to migrate. Physical vs logical as you kind of touched on. Yet still to this day we have companies deploying entire code bases to multiple regions AND disaster recovery sites in active, active, passive setups AND still requiring antiquated tape backups AND now with the added bonus of paying exorbitant premiums for "virtualization" of cloud based hardware and services that always have to be on so they can sync but somehow dont count as backups. And new white papers and nomenclature and buzz words furthering the grift every year.

Maybe I was being coy, or perhaps obtuse?
So, do answer this with a substantive answer or with the kneejerk answer? My first reaction after reading this:

HB21iJuWoAAAMD7.jpg



Virtualization was going to make it easier to migrate from hardware to hardware, and guess what? IT IS. VMware has saved everyone thousands and thousands of hours moving guests around hosts in the early days, and now everyone has Azure and AWS accounts that can spin up servers and install apps at the drop of a hat. You can have a server built for pennies in minutes that would have taken days on hardware that would have cost 10-100s of thousands. I managed a Data Center. While nothing beats racking a new 3U server over my head without a lift, virtualization has made it so many businesses don't need or want to be in the DC business any longer. Data Centers have consolidated, X As A Service has grown exponentially: Platform aaS, Compute aaS, Software and Infrastructure obviously, it's a huge list.

EDIT:An added benefit that I directly see in my line of work is the number of Major Incidents we used to see raised from crashed servers brought off line by hardware failure has been dramatically reduced. Every critical app, no matter how well architected, even those with big iron servers with HA pairs, etc, suffered when inevitability struck and a hard drive went down or a fan failed. RAID arrays suffered enough losses and then BOOM. Now that has been moved to large SAN clusters, guests shuffled around on huge hosts or in massive Azure or AWS farms which we pay for as a service rather than having direct control (or have to maintain), 5 9's is an expected SLA that is readily achieved. THAT is a promise on which virtualization handily delivered as well, once applied at scale.

I don't think "obtuse" captures the nature of our disconnect.

I really don't get ultimately what you are complaining about: Is it that people/businesses feel compelled to do more now than they used to in the past as technology extends capabilities, or that the prices of this layering is more expensive than you think it should be, or that people are adding terms to refer to different technologies and strategies as they come into use?

Help a brother out.
 
Last edited:
  • 1Like
Reactions: 1 user

Khane

Got something right about marriage
21,900
15,872
It's 2010. I have a simple application that is business critical but doesn't see much traffic or load. I deploy it to a VM. I would like high availability so I also deploy it to another, passive VM and take advantage of clustering. In the year 2010 we call this redundancy and refer to the passive VM as a backup. But we also take snapshots of the VM at regular intervals (backups in the more literal sense) so we can stand it back up quickly in case of catastrophic failure (both servers go kaput). We don't even really need to have the second VM, because of how virtualization works, we do anyway because of service agreements to ensure 99.9% uptime instead of ~95% uptime.

It's 2026, marketing and corporate greed has changed what words mean. We have CI/CD now! It's amazing. We don't own or manage our own hardware anymore, megalithic corporations do that for us now for a small fee. They have ensured us they are resilient and can stand up new servers for us in the blink of an eye, again, for the right price. Everything is virtualized but we no longer actually take advantage of any of that shit. We have our pipelines deploying to even more always on environments for no fucking reason at all other than we've bought into the grift hook line and sinker. We even keep our disaster recovery environment up at all times now because... standards and best practices or some shit! I don't think we even take snapshots anymore, or at least our cloud provider maybe does? Who knows? One thing is for sure, none of these supposedly identical environments are considered backups, because fuck the English language if it gets in the way of profits.

I don't think you and I are disagreeing on the what, more the how and the why. A lot of times the purpose of a particular type of technology gets lost in the weeds or suppressed because of money.

A VM should be portable, nobody actually uses that portability anymore. Not really.
 

Xerge

<Donor>
1,564
1,703
Just an update:

  • Closing in 4k gross and subscriber count has increased.
  • Word of mouth spread of app
  • haven't written a single line of code
  • All income from app has paid for AI usage
  • Github copilot still remains top dog in my work flow.
  • AI can do bulk/volume work a human never will; its comical the amount of output an AI can really do.
  • Solved weekend todo lists in a single evening, easily.
  • AI loves blueprints and spec docs. Chrome extension>windows app>and now android app shows clear progress and iteration.
  • Latest AI is in fact trained well on commonly known algos, principals, etc. I did a volume normalization feature for Exoplayer on Android in a couple of hours using the well known MIT algo. 1 shot, boom, done.
  • A customer had an IPTV playlist that was geo blocked/managed; I gave claude the situation, I directed it via prompts on prep work, and boom 1 shot. Customer in france can can access and playback their content.

I could go on and on, but honestly.. this shit is fucking real. There will be a societal divide; can't use AI? get fucked. Got no AI sub? get fucked on, youre a loser.
 
  • 1Like
Reactions: 1 user

Xerge

<Donor>
1,564
1,703

Is iCloud "top dog"? Short answer: No, not in the way that matters for what we're building.

iCloud/CloudKit is dominant by install base (every Apple device) and end-user UX (truly invisible). But by architecture and developer experience it's middle-of-the-pack. The interesting frontier is elsewhere.

The landscape, ranked by axis​

1. Invisible end-user UX
  1. iCloud / CloudKit / CKSyncEngine — won't be beaten on Apple platforms. Zero-config from the user's POV.
  2. WhatsApp (E2E + multi-device, custom protocol on top of Signal)
  3. Google Photos / Drive
2. Modern architecture closest to what we're doing
  1. Replicache / Reflect (Rocicorp) — event log + mutation IDs + server-assigned order + client-side rebase. This is structurally identical to XMP sync v2. If you want to read one company's blog to validate our design, it's theirs. Aaron Boodman (ex-Dropbox, ex-Google Gears) literally formalized the pattern.
  2. Linear's sync engine — custom, WebSocket-based, event log, mutation IDs, server timestamps, client rebase. Famous for feeling instant. Tony Dipp's talks describe it; same primitives as ours.
  3. Figma multiplayer — custom CRDT-ish (their tree CRDT for layers). The Evan Wallace blog post is canonical.
  4. Notion — block-level ops with server-assigned ordering. Mostly proprietary, similar shape to Linear.
  5. Local-first (Ink & Switch 2019 manifesto + Automerge) — the intellectual home of the modern sync movement.
3. CRDT correctness (mathematical merge guarantees)
  1. Yjs (rich text, collab editing) — production-grade. Used by Liveblocks, Hocuspocus.
  2. Automerge — general JSON CRDT. Heavier wire format than Yjs.
  3. ElectricSQL — Postgres ↔ SQLite active-active with CRDT semantics.
4. Dev API ergonomics
  1. Firestore — realtime listeners + offline persistence + transactions. Easy to ship, but conflict resolution is server-timestamp LWW (no real merge).
  2. Realm Sync / MongoDB Atlas Device Sync — object-level sync with rules. MongoDB sunset Device Sync Sept 2025, so write it off for new work.
  3. Supabase Realtime — Postgres logical replication over WebSocket. Less complete than Firestore but open-source.
  4. Couchbase Lite / PouchDB / CouchDB — MVCC rev-tree replication; multi-master. Mature but the rev-tree merge model is unintuitive.
5. File-level sync at petabyte scale
  1. Dropbox (Magic Pocket, content-defined chunking)
  2. Google Drive
  3. iCloud Drive

Where CloudKit specifically falls short​

  • Apple-only. Hard veto for any cross-platform product.
  • Schema rigid. CKRecord types are essentially typed dictionaries; changing them is painful.
  • Opaque debugging. When sync stalls there's almost no diagnostic surface. Famous pain point for indie devs.
  • Conflict model is LWW with server-record-on-conflict. No CRDT, no event log. Apps have to roll their own merge logic on top.
  • CKSyncEngine (iOS 17+) finally gave Apple devs delta sync with change tokens — which is what Firestore had in 2016 and CouchDB had in 2005. It's catch-up, not leadership.

How XMP sync v2 stacks up​

Our design = Replicache-shape, slimmer:
PropertyXMP sync v2ReplicacheCloudKitFirestoreAutomerge
Event log✓ (DO ev: ring)✗ (record store)✗ (doc store)✓ (op log)
Server-assigned monotonic seq✓ (change tokens)✓ (server ts)
Per-mutation idempotency key✓ ((deviceId, mutationId))✓ (mutationID)~ (record etag)n/a
Lamport / causal ordering~✓ (vector clocks)
Optimistic local apply(P2)
Rebase on pull(P3)✓ (merge)
Cross-platform


Conclusion: the architecture we're shipping is the modern, defensible one — same family as Replicache, Linear, and the local-first movement. CloudKit is not the model to chase. If you want a single name to study for cross-check on your design choices, it's Replicache (Rocicorp's blog + their "How Replicache Works" doc). For inspiration on the UX bar you're aiming for, Linear. For the philosophical north star, the Ink & Switch local-first essay.

Round 2 review summary​

  • 1 real fix applied: drainOnce now self-reschedules on full-batch sends to drain backlog instead of waiting for the next emit(). G1 BUILD SUCCESSFUL (8s).
  • 4 design constraints documented in §13a (cursor regression idempotency requirement, hasClientId O(N), two-put pattern, lamport-id race, manifest schemaVer).
  • No security or correctness issues in the worker on second pass.
  • G2 wrangler deploy still gated by Cloudflare incident znll7lfnd35h. Standing down on retries — ping when status flips.