Mike G.

26.9K posts

Mike G. banner
Mike G.

Mike G.

@MikeKGilmore

Husband, Father of five. Software Engineer. Liberty enthusiast. Lover of Faith, Family, and the Constitution. Utah. #utpol Views are my own

Salt Lake City, UT Beigetreten Ekim 2012
761 Folgt960 Follower
Mike G.
Mike G.@MikeKGilmore·
@Bensam123TV @straceX I honestly hope you're right, but as I do my own cost forecasting, I can't be this optimistic. When subsidy ends (which it must) it will be quite rough and the industry will likely stratify (only elite tech firms/govt will have the best models).
English
0
0
0
2
Sam Adams
Sam Adams@Bensam123TV·
@MikeKGilmore @straceX This doesn't take into account the rate of improvements in hardware, expansion of data centers, or improvements in models including efficiency. All of which are ballooning wildly. Pretending this is the worst it's going to be is disingenious and silly.
English
1
0
0
35
Gracia
Gracia@straceX·
this is the most satisfying plot twist in tech history
Gracia tweet media
English
75
3K
36.5K
450.5K
Mike G.
Mike G.@MikeKGilmore·
@Bensam123TV @straceX Tokens are 90% subsidized right now (10x) on current models. Next-gen 10T models like Mythos are so expensive they can't be run at scale. They would need 30x-50x (or even much more) subsidy on current subscription tiers.
English
1
0
0
52
Sam Adams
Sam Adams@Bensam123TV·
@straceX I bet.... Today on our copium LARP. Maybe if you're hiring highly skilled developers out of Africa. Its getting tough to justify costs of human labor and that's with tools we have today. Given it another year and it's going to be very difficult.
English
5
0
4
3.2K
Mike G.
Mike G.@MikeKGilmore·
@kareem_carr When the token subsidy ends it will k*ll-shot vibe-coded startups AI will enable speed but that will not necessarily always translate to profitability
English
0
0
0
10
Mike G.
Mike G.@MikeKGilmore·
@kareem_carr Yes - current models are subsidized at 90% (10x cost). Future models such as 10T will be 30-50x cost. There will be no other developments past 20-30T because they will be too expensive to run
English
1
0
0
128
Dr Kareem Carr
Dr Kareem Carr@kareem_carr·
so let me see if i got this AI stuff straight: - you pay to experiment with the models - you pay to automate your business workflows - you pay to validate the outputs of said workflows - you pay to detect any silent downgrades in the model - you pay to update your workflows - you pay to revalidate your workflow outputs - you pay to do it all again every time there's a major model update - you pay 10x all that when they finally stop subsidizing the models and you have to pay the real price - also you don't even get any business advantage out of it because literally anybody can pay for the exact same AI you can i'm no business genius but this sounds like a terrible business model for everybody except the AI companies.
English
99
124
946
47.4K
Mike G.
Mike G.@MikeKGilmore·
@xwanyex For us, we're getting through our backlog about 33% faster or so. We're rolling out an experimental fully-featured ADLC workflow (agentic software development), but I predict these workflows will be too expensive to run when the token subsidies end in the next couple years.
English
0
0
1
360
wanye
wanye@xwanyex·
I don’t have to be convinced that LLM’s make programmers more productive. But where’s all the stuff? We’ve now had months and months of 100x or 1000x programmet productivity improvements. Where’s all the stuff they’re building?
English
636
167
7.3K
564.7K
Mike G.
Mike G.@MikeKGilmore·
@CtrlAltDwayne Possibly all for the better - when token subsidy ends it will likely kill-shot most vibe-coded startups
English
0
0
0
42
Dwayne
Dwayne@CtrlAltDwayne·
The vibecoders are starting to give up because they're learning AI can't do everything for you. They're tools that are powerful in the hands of people who can code, but vibecoders are not real devs.
Dwayne tweet media
English
390
250
2.9K
167K
Mike G.
Mike G.@MikeKGilmore·
@bitforth @OmariGaskinsJr @CtrlAltDwayne One hundred percent. A well-structured monorepo is a sign of a very mature engineering org. Also, many small repos is a sign of poor architectural oversight or siloing (there are exceptions - for example, IaaS, etc)
English
0
0
0
21
Alan
Alan@bitforth·
@OmariGaskinsJr @CtrlAltDwayne Some of the best engineering orgs on the planet run on massive monorepos. It’s usually far easier to operate one well-structured monorepo than babysit 50 scattered repos that constantly break each other through version drift and cross-repo dependencies.
English
1
0
3
128
Mike G.
Mike G.@MikeKGilmore·
@slimjimmy @GrizzledTexan Yep it makes sense that since LLM is a language model, it would be excellent at syntax. I think Mixture-of-Experts will offer some promising results in the space of reasoning and software architecture, but I think you're mostly correct that this will have limited results
English
0
0
2
128
Slim Jimmy
Slim Jimmy@slimjimmy·
@GrizzledTexan there's a difference between software engineering and coding, of which the latter is a small part, hence marginal
English
7
2
260
9.8K
Slim Jimmy
Slim Jimmy@slimjimmy·
where i'm at on LLMs: 1. LLMs are NOT going to outmode software engineers, not now, not ever 2. LLMs alone are a dead end for attaining anything like AGI 3. forget about LLMs achieving ASI 4. LLMs will only make good engineers faster, marginally 5. LLMs will remain poor at architecture and design this is 100% because LLMs cannot reason and no, generating a bunch of hidden context is not "reasoning" as always, when the same people are making claims like this, have a look at what they stand to gain and to them, i warn directly: you are going to fall. hard
English
362
483
4.9K
289.1K
Mike G.
Mike G.@MikeKGilmore·
@TB1Kinobe The engineers aren't the problem. It's the investors who are buying up huge tracts of land with the express purpose of packing in as many housing units as possible.
English
0
0
0
83
TB1™️ 🇺🇲
TB1™️ 🇺🇲@TB1Kinobe·
25 years ago after engineering school I was involved in a residential construction project where speed was the focus. I did not know the scale or how widespread it would get. None of us did We were young. I have been out of it for over 10 years now with the deepest regrets of contributing to this process at a high level. I don't care if anyone sees this or even gets one like I just wanted to say I'm sorry.
English
379
99
2.6K
604.7K
Mike G.
Mike G.@MikeKGilmore·
@gnukeith If anything it's evidence that there's a token squeeze going on which is limiting OC usefulness or practicality. I've only seen nothing but very poor token efficiency from OC workflows.
English
0
0
1
485
Mike G.
Mike G.@MikeKGilmore·
@patomolina I think something worse is going to happen in the next year or two: Currently tokens are heavily subsidized (by up to 90%). When they are no longer, it will kill entire businesses built on vibe-coded apps. Others will no longer be able to afford AI at scale
English
0
0
0
17
Pato Molina
Pato Molina@patomolina·
Anthropic decidió dar de baja a toda nuestra organización por una supuesta infracción de sus condiciones de uso. Qué política específica infringimos no tengo ni la menor idea: simplemente recibimos un mail y listo, adiós Claude. Si querés apelar la medida hay que completar un Google Form, así de ridículo como suena. De golpe más de 60 personas se quedaron sin una herramienta fundamental para trabajar. Integraciones, skills, historial de conversaciones: todo perdido o, en el mejor de los casos, parado por tiempo indeterminado. Enorme aprendizaje para cualquier empresa de software que dependa de herramientas de IA en procesos críticos. Nunca hay que poner todos los huevos en una canasta.
Pato Molina@patomolina

@claudeai you took down our entire organization with 60+ accounts belonging to a legitimate company for no apparent reason, without any explanations. The only way to appeal the decision is by filling out a Google Form? Very bad UX and customer service.

Español
790
1.4K
9.5K
5.1M
Battle Programmer Yuu
Battle Programmer Yuu@netspooky·
Is there a term for when people spend exponentially more time trying to get an LLM to do a task correctly, than it would have taken to just do the task themselves?
English
230
22
743
43.3K
Mike G.
Mike G.@MikeKGilmore·
@aphysicist As production capacity approaches infinity, the cost of production (and the value of currency) approaches zero This assumes that production is not stratified (a big assumption)
English
0
0
0
14
Mike G.
Mike G.@MikeKGilmore·
@TomJBelfort @Jason______A @DaveRamsey Yup - they basically just gifted that principal amount to the bank. I'm in a low-3% loan on mine - the bank was practically begging me to refi a couple years ago. I told them I'm gonna ride this pony 'til it dies
English
1
0
3
267
Tom Belfort
Tom Belfort@TomJBelfort·
It's worse than that, Jason. He didn't just pay it down and keep paying the same monthly payment. He reduced his monthly payments by $1400/mo, paying the same total interest. All he's done is taken a loan of $1400/mo from his own money, where he makes no interest on it. He would have been better off putting it in the bank an earning 4% and taking $1400/mo
English
8
0
49
10.2K
Jason Applebaum
Jason Applebaum@Jason______A·
Whenever I see things like these I assume it’s rage bait but then I remember how stupid the average American is and I understand why @DaveRamsey is so wealthy. Imagine paying down a 3.75% mortgage. The bank jumped for joy. They prob had a board meeting to party after seeing that $260k gift.
Average Money Joe MD@averagemoneymd

Last month, we received a family gift of $260,000. We debated a lot and ultimately decided for a few reasons to drop it all on our 3.75% mortgage. Our mortgage went from $750k --> $490k We recognize we could have made much more investing, but it felt secure to take a little risk off the table. Mortgage calculator shows this will save us $285k in interest. We then recasted our mortgage and our payment dropped from ~$3,800 to ~$2,400. A little bit closer to freedom.

English
115
46
1.7K
920.5K
Mike G.
Mike G.@MikeKGilmore·
@ITARviolation I haven't run the numbers but it looks like they refi'd into a new 30y - looking to 'reduce monthly expenses' So instead of continuing the same monthly payment to get ahead of the high interest period, they opted to dump probably close to that principal amount into.. interest
English
0
0
1
2.5K
addison raetheon
addison raetheon@ITARviolation·
this might be the dumbest financial move I’ve heard of all year
English
14
2
1.8K
140.9K
addison raetheon
addison raetheon@ITARviolation·
spent… $260k… paying down a mortgage… at 3.75%…
addison raetheon tweet media
Average Money Joe MD@averagemoneymd

Last month, we received a family gift of $260,000. We debated a lot and ultimately decided for a few reasons to drop it all on our 3.75% mortgage. Our mortgage went from $750k --> $490k We recognize we could have made much more investing, but it felt secure to take a little risk off the table. Mortgage calculator shows this will save us $285k in interest. We then recasted our mortgage and our payment dropped from ~$3,800 to ~$2,400. A little bit closer to freedom.

English
151
238
17.3K
3M
Mike G.
Mike G.@MikeKGilmore·
@KevinNaughtonJr It's not. Anthropic can't afford general-access inference on Mythos on their current subscription model (actually likely no AI company can afford inference on any 10T model). Only the big enterprises will get access to it. The stratified world begins with 10T
English
0
0
1
51
himanshu
himanshu@himanshustwts·
@RedwoodFounders i am interested in post IPO dynamics lol, kinda bearish on ant
English
1
0
0
1.1K
himanshu
himanshu@himanshustwts·
few cents on anthropic and openai launches today (and from quite sometime). i am almost convinced that anthropic has figured out the game. essentially they are leveraging consistent distribution + public perception + effective marketing and narrative compounding. (irrespective of kind of launch, you can track this from months) and openai has been surprising me with this from long time. they are coming up with more exciting stuffs (see computer use launch today) but aren't consistent. they come up with (extra) news and lores which makes it distributed. we have been carrying lot of hopes from anthropic to see something big EVERY day and it has been rewarding for them.
English
10
3
133
25.6K
Mike G.
Mike G.@MikeKGilmore·
@JakeKAllDay @P33RL3SS @VictorTaelin Yeah - but may have backfired this time. They got the US Treasury and Federal Reserve up in arms due to their rhetoric. I think they will have to tamp it down or the public is going to turn on them very quickly
English
1
0
0
14
Jake
Jake@JakeKAllDay·
@P33RL3SS @MikeKGilmore @VictorTaelin It probably is a security risk but so was gpt 5.4 pro if you tried hard enough. But it’s the same shtick Amodei ran with GPT2– the thing you cant serve at full scale you say you *wont* bc it’s too powerful /justify scarcity. It’s amazing marketing!
English
2
0
1
38
Taelin
Taelin@VictorTaelin·
Anthropic just launched a model that loses in ALL benchmarks to another model... that is also theirs 😭 I'm thankful yet worried - are we entering a world where the big labs just stop publishing their leading AI's? As a Brazilian I can only ask: for the betas, what remains?
Claude@claudeai

Introducing Claude Opus 4.7, our most capable Opus model yet. It handles long-running tasks with more rigor, follows instructions more precisely, and verifies its own outputs before reporting back. You can hand off your hardest work with less supervision.

English
168
26
1.2K
113.1K
Mike G.
Mike G.@MikeKGilmore·
@JakeKAllDay @VictorTaelin I think this is correct, Anthropic can't afford to run inference on a 10T model via the current subscription model. Estimated cost is $20/MTok They'll market this to the huge enterprises and release a mini version to the general public.
English
1
0
1
37
Jake
Jake@JakeKAllDay·
@VictorTaelin It’s vs a model 10x its size though, in potentially a dense architecture. Models that size are still experimental tools. Same as GPT 5.4 pro vs 5.4. Anthropic is never going to release a model that size as part of the main subscriptions. Just gotta wait 6 months for the distill!
English
2
0
1
953

Entdecken

@Bensam123TV @straceX @kareem_carr @xwanyex @CtrlAltDwayne @bitforth @OmariGaskinsJr @slimjimmy