Mike Kelley

Public Key

npub1d3yy5ny3jujefyar0a33wtupahusmfwftpc4ca4zpdk2a4fahvxqyf0n0v

NIP-05 Address

Profile Code

nprofile1qqsxcjz2fjgewfv5jw3h7cch97q7m7gd5hy4su2uw63qkm9w657mkrqpp4mhxue69uhkummn9ekx7mqprpmhxue69uhhyetvv9ujumt0d4hhxarj9ecxjmntgjz5jn

Publishing to

Author Public Key

npub1d3yy5ny3jujefyar0a33wtupahusmfwftpc4ca4zpdk2a4fahvxqyf0n0v

Show more details

Last Notes

2026-04-24 00:49:38 UTC

- reply

otherwise I'm listing it on FaceBook marketplace next week. I replaced it with two r620 smaller machines and I just don't have the need for the beast anymore.

2026-04-24 00:48:09 UTC

@npub1rcn…dw4h @npub1fmx…wp74 I just finished listening to the last episode... at the end you were talking about VM costs and upsizing the VM.
I've got a Dell r720XDS with 12 4TB SAS drives in a RAID-10 configuration. 96GB RAM, dual 10 core Xeon processors and dual 1000k power supplies. Runs great, clean and came out of a datacenter.
Datacenter costs is $160 per month in a Tier-3 center.
Looking to get $700 for the server. Let me know if you want to jump to bare metal!

2026-04-23 12:37:20 UTC

- reply

you lost me on all this. Bring me up to speed on what you're working on and what was this in regards to?

2026-04-16 23:10:27 UTC

- reply

sounds like a do while 1=1 loop in someone's code....

2026-04-14 21:35:54 UTC

- reply

is this the shut down that I keep seeing people complaining about on Facebook?

2026-04-14 21:32:53 UTC

- reply

"DON'T EAT YELLOW SNOW!" -- Things you should know.

2026-04-14 21:25:26 UTC

- reply

The 24 hour feed report is now carrying the extra information as well. It's basing it's determination on the LLM analysis if available instead of just keyword heuristics.
What does this mean?

2026-04-13 00:31:25 UTC

- reply

it can be. I typically like to choose servers that are located in South Korea, Japan, Canada or the United States. I definitely stay away from those locating in China. And I haven't even tried any from Vietnam. The one thing I do like is the reputation rating.

2026-04-08 12:02:24 UTC

- reply

have you looked at vast.ai? The pricing for their 5090s are significantly lower than runpod.ai
But if you're just doing transcription of an audio podcast, why aren't you using assembly.ai? A 2-hour podcast probably would cost about $1 to transcribe in under a few minutes, and you wouldn't have to spin up any virtual machine -- it's just a simple API call that returns back a json payload.

2026-04-08 11:55:10 UTC

- reply

that was my biggest concern. Was the context window only being 128K, although after this week I've seen quite a few videos of people talking about the new Gemma 4 that supports 256k and it will run on a 3090.

2026-04-08 00:38:02 UTC

- reply

also, another question regarding your RTX 3090... What are you thinking about that? From what I heard, that's pretty much The Sweet spot on gpus. What's your feelings regarding that particular model? Are you finding the 24 GB enough?

2026-04-08 00:36:07 UTC

- reply

I'll have to give fish a try. I went out and checked it out on the website and it looks interesting. It shares some of the aspects that qwen TTS has that I think beats Google Gemini TTS. I've got a specific project in mind that I want use to compare the two different TTS types and see which one performs best in quality.

2026-04-06 19:41:26 UTC

- reply

have you used Qwen TTS? It blows away Eleven Labs! I use it on my small GTX 1060 6GB card and it produces speech comparable to Gemini TTS.

2026-04-02 04:57:45 UTC

"The revolution will not be downloaded."
"The revolution will not be ad-read."
"The revolution will not be a limited series."
"The revolution will not be in your feed."
"The revolution will not be IAB compliant."
"The revolution will not be a two-mic interview."
"The revolution will not be a limited-run true crime series."
"The revolution will not be played at 1.5x speed."
"The revolution will not be a Spotify Exclusive."

2026-03-26 16:57:23 UTC

- reply

I would suspect the reason you're seeing higher toking usage in Claude is because the the maximum context window for gpt 5.3 is 400k where the maximum contacts window of Claude 4.6 is 1 million. Therefore, it has the ability to send a whole lot more information per prompt request. The beast is hungry and it loves to be fed!

2026-03-20 00:55:44 UTC

- reply

did you see that crypto, gold, and silver all tanked today.

2026-03-18 03:20:30 UTC

- reply

do you host your server yourself or do you use a third party service like Masto.host or SpaceBear? And if you do host it, do you have to manage it? And how much time do you spend managing it if you do?

2026-03-17 19:32:07 UTC

@npub1rcn…dw4h are you storing episode data or just podcast data?

2026-03-17 14:52:31 UTC

@npub1fmx…wp74 what is/are the advantages of setting up my own mastodon server vs just posting on others such as yours?
I was considering setting one up, but I'm not quite sure why I should...

2026-03-16 19:16:38 UTC

- reply

yeah I can see what you're saying... Pretty much every podcast in those categories would be marked true. 😂

2026-03-16 18:02:38 UTC

Can I suggest a new tag for podcast index specifically for business, finance, self-help, and entrepreneur podcasts.
Just a simple tag that indicates whether it is a contrepreneur format. True or False value.
Does anybody remember the old service Klout that existed about 10 years ago or so... That was actually a pretty good concept.

2026-03-11 19:46:31 UTC

- reply

Coming Soon...
Inspired by the "fog of war," The Fog of AI is your reconnaissance for the synthetic era. In a landscape of deepfakes, black-box algorithms, and rapid disruption, we cut through the digital haze to provide clarity. We track the truth, explain the tech, and map the ethics of an unpredictable future. Don’t get lost in the hype—lift the fog and navigate the AI frontier with us.

2026-03-11 18:31:47 UTC

- reply

oh the, this won't stay free forever is already starting to happen... I upgraded my Google workspace account just so I could get the new added skills for their AI studio just to find out that they've removed them and moved them up to Enterprise level pricing.

2026-03-11 17:11:51 UTC

- reply

LOL! I worked with a startup that did this exact same thing about 6 years ago and was acquired by Spotify. Spotify currently holds the patents on the technology.

2026-03-11 12:34:48 UTC

- reply

yes. The problem that arises is the same concept of making a copy of a copy of a copy of a copy of a copy of a copy of a copy of a copy of a VHS tape. Every time you make a copy it deteriorates. Ultimately you get junk!

2026-03-11 03:56:37 UTC

- reply

another great source which can be a little annoying at times is openrouter.ai if you have a $10 balance on account, then you can use a very generous allocation of their free models. I went ahead and put $10 on my account a few months ago and I use the free models quite a bit for small tasks.

2026-03-11 03:49:45 UTC

- reply

I don't have a problem with AI generated podcast content. Whether it be the actual content itself or the spoken audio that is generated. What I do have a problem with is the integrity of the content. A small little story was buried last week in AI news and that there are two very large llm providers that are training their model based on responses from the very large LLMs from the big players. This is a very major problem!

2026-03-11 03:45:20 UTC

- reply

it's interesting that you posed the question whether or not you should spend $20,000 to run your own models locally. I've been asking myself that exact same question the last few weeks, specifically whether or not I should spend $500, $1,500, $3,000, or more?
What I came up with was that the smart move right now is to stick with my limited 6 GB GPU, use my Google workspace subscription, use my Google cloud free tier availability, and lease time on vast ai