Steve Bain

Development, Engineering, Coding.

[Steve Bain @g2sb]

@cremieuxrecueil Some poetic license in X posts is accepted, but the comparisons are 10-15mg of tirz vs 7.2mg of sema and neither are "tons". Of course, tirz wins easily in side effects and maturity at those doses.

[Steve Bain @g2sb]

@Braders2k @Teknium @tonbistudio https://t.co/iE2OwXNQGy

[Steve Bain @g2sb]

@JRK_was_here @DrSeanOMara In my research so far, the answer is "nowhere".

[Steve Bain @g2sb]

@BraveSampson @tread92 @BrendanEich @rebron On Windows, the built-in Zira voice is pretty darn good and low-resource. Works great in the reader modes of both Brave and Firefox. We need a good low-resource voice for Brave on Linux.

[Steve Bain @g2sb]

@brah_ddah @ErikVoorhees @AskVenice Check the pricing page:

https://t.co/mnZA9GxPbs

[Steve Bain @g2sb]

@FireCleanse @ErikVoorhees If you have use for the API, there is good reason to mint Diem with your staked VVV. (Make sure to keep 100 staked VVV for a Venice Pro subscription.) You can always burn the same Diem to get your staked VVV back, regardless of current token prices.

[Steve Bain @g2sb]

@AJsolspace @solanamobile @SeekerClaw You can buy $5 of API credits on https://t.co/rjLLwLydng and the select the Haiku model at install. That will give you enough compute to play, and experimenting is most of what SeekerClaw is good for at the moment.

[Steve Bain @g2sb]

@matt_elevenlabs 11B

[Steve Bain @g2sb]

@cherrydotfun grafman.skr

[Steve Bain @g2sb]

@FUTURESaiLOR_AI @ErikVoorhees @openclaw MiniMax M2.1 is very good at about half the cost of Kimi K2.5, but without the multi-modal functionality. (Kimi is great at image understanding.)

[Steve Bain @g2sb]

@ErikVoorhees @openclaw Kimi K2.5 is marked as anonymized, not private. That is an important distinction. Will Venice soon run Kimi K2.5 on its own infrastructure?

[Steve Bain @g2sb]

@simoneDotDev @ErikVoorhees Was it ever unbroken? I added the JSON addenda in the PR discussion and that worked. Venice Claude 4.5 Opus is not fast, but very competent. Unfortunately burns thru Diem credits like wildfire. I suspect that entire context history is being sent with each call.

[Steve Bain @g2sb]

@petrrosslerjr @naomibrockwell @brave @AskVenice In @Brave, you can subscribe to better models or "bring your own" model from any OpenAI compatible API you choose. Works well for me with Venice (for private or anonymized), Perplexity, OpenRouter, and local models.

[Steve Bain @g2sb]

@aerosol_xyz @burnandclaim grafman.skr

[Steve Bain @g2sb]

@solanagaming @solanamobile @burnandclaim grafman@skr

[Steve Bain @g2sb]

@burnandclaim grafman.skr

[Steve Bain @g2sb]

@tucker4115 @teslayoda @BowTiedphrma @elonmusk @intel Ahh. Thanks. But I'm not sure how *that* Fremont experience is relevant to possible teaming with an existing player in creating a greenfield fab.

[Steve Bain @g2sb]

@teslayoda @BowTiedphrma @elonmusk @intel Is there a reference for "learnings from Fremont"?

[Steve Bain @g2sb]

@eh_hmmm @AskVenice Something I didn't understand until very recently: If you are staking VVV there is little risk to minting DIEM except 20% tax on staking rewards. You can always burn the same DIEM for the same VVV regardless of variations in market or minting rates.

[Steve Bain @g2sb]

@eh_hmmm @AskVenice Apparently, staking 100 VVV now merits a "Pro" subscription as an alternative to annual $$. But also, of course, a "Pro" subscription has become less distinctive from unpaid as many/most of the more interesting model options are now metered.