Mike Kelley

Public Key

npub1d3yy5ny3jujefyar0a33wtupahusmfwftpc4ca4zpdk2a4fahvxqyf0n0v

NIP-05 Address

Profile Code

nprofile1qqsxcjz2fjgewfv5jw3h7cch97q7m7gd5hy4su2uw63qkm9w657mkrqpz3mhxue69uhhyetvv9ujuerpd46hxtnfduqs6amnwvaz7tmwdaejumr0ds4dklvu

Publishing to

offchain.pub

relay.damus.io

nos.lol

Author Public Key

npub1d3yy5ny3jujefyar0a33wtupahusmfwftpc4ca4zpdk2a4fahvxqyf0n0v

Show more details

Last Notes

2026-10-08 16:35:49 GMT

- reply

I agree. I just think that it's going to evolve into the same situation that we see with ad blockers for websites and YouTube. It becomes a cat and mouse game. It works to some degree, some of the time, with a lot of patience, and ultimately just for a small minority of people.
But anyhow, I enjoy having conversations on topics like this. There's no right or wrong answers.

2026-10-08 16:22:38 GMT

- reply

but let's also consider this from a different angle. Let's assume that we live in a perfect world and we can 100% of the time easily remove dynamic ads from a podcast and it's 100% accurate and perfect.
Is it ethical? Advertisements are put in place to pay for the production of the podcast and its content. Podcasters make the podcast available for free to listeners in exchange for the listener to be subjected to advertising in the content. Is removing dynamic ads stealing?

2026-10-08 16:08:46 GMT

- reply

oh it is possible. But right now it involves two steps. First transcribing and then using the llm to find the advertisements. Where it becomes impractical is whether the listener is willing to wait 5 to 10 minutes to process an audio file to determine the ad breaks for removal on playback. And performing these tasks requires power which drains battery time. You multiply this by the 5 to 10 podcasts that the listener subscribes to and the reality starts to sink in...

2026-10-08 15:57:53 GMT

- reply

and you are correct that DAI can be trashy and disruptive... But that is because the producer of the podcast allows it to be. Once listener numbers drop off, producers will find the right balance that matches the acceptable terms for the listener and at the same time meets the requirements of the producer and the advertiser.
I personally find 4x 15 second batched advertisements a whole lot more effective and less disruptive than a single 1-minute advertisement.

2026-10-08 15:49:04 GMT

- reply

removal of dynamic ads at first glance seems quite easy, but it becomes complicated and complex quickly. It comes down to is timing. You must know the exact time for ad breaks, length, and finally when the break ends. Dynamic ads served on a per individual basis can cause unique timing issues where all the advertisement details fluctuate for each user. So ultimately, the podcast app would have to rely on processing the audio file on the device itself. Which isn't practical today.

2026-08-31 22:47:45 GMT

- reply

yeah, I don't think people will quite understand that when they use cloudflare they end up putting themselves behind a gated wall whether it be a website or podcast hosting. The concept and the idea that they provide has good intentions. But ultimately when you start blocking traffic it doesn't always work as desired.

2026-04-24 00:49:38 GMT

- 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 GMT

@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 GMT

- 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 GMT

- reply

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

2026-04-14 21:35:54 GMT

- reply

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

2026-04-14 21:32:53 GMT

- reply

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

2026-04-14 21:25:26 GMT

- 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 GMT

- 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 GMT

- 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 GMT

- 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 GMT

- 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 GMT

- 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 GMT

- 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 GMT

"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 GMT

- 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 GMT

- reply

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

2026-03-18 03:20:30 GMT

- 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 GMT

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

2026-03-17 14:52:31 GMT

@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 GMT

- 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 GMT

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 GMT

- 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 GMT

- 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 GMT

- 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 GMT

- 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 GMT

- 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 GMT

- 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 GMT

- 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