godwin. _10
@CodingGodwin
Followers
70
Following
651
Media
44
Statuses
1K
Multimodal LLMs · Vision + audio pipelines · RAG · Style transfer · Mirror-neuron architectures · Narrative modeling ·
Joined May 2023
We just raised $16M. Introducing Arcads 2.0, the AI ad maker that outperforms humans. RT and comment “Editor”, I’ll send you 1,000 AI trained editors for hire 👇
2K
1K
5K
The emotional graph of a founder is just as volatile as a crypto chart.
2
1
10
Data poisoning isn’t an arms race we’re trying to win with bigger models or detectors. It’s a war we’re opting out of by never stepping onto the battlefield in the first place. Healthy virality > maximum virality.
0
0
0
250 poisoned samples can kill a 70B model trained on everything. 250 poisoned samples can’t even get past our ingestion filter when the entire dataset is 0.1 % of the internet and every category is explicitly blessed.
1
0
0
We’re now training Jarvimind on content that spreads because it genuinely lifts people up .....from inovation and founder stories,explainers that make something simple to understand,stoic wisdom and so much more stuff that will improve humanity
1
0
0
That’s exactly why we pivoted Jarvimind last week. Old plan: 1B+ random viral clips → trivial to poison. New plan: train only on a tiny, human-vetted slice of genuinely valuable content.
1
0
0
The only reliable defense is radical reduction of the attack surface. If your training data is “everything that ever went viral”, you’re playing whack-a-mole forever.
1
0
0
No model trained on the open internet will ever be fully safe. The new Anthropic poisoning paper shows that ~250 malicious samples are enough to backdoor LLMs of any size — permanently. Model scale gives zero protection.
1
0
0
Comment with your github
If you’re into the mission and love working with open-source well in search of a technical co-founder — comment below or DM your GitHub and we’ll see the way forward.
0
0
0
If you’re into the mission and love working with open-source well in search of a technical co-founder — comment below or DM your GitHub and we’ll see the way forward.
0
1
0
If you’re into the mission and love working with open-source well in search of a technical co-founder — comment below or DM your GitHub and we’ll see the way forward.
0
1
0
WE Don't promote: content that encourages addictive scrolling content designed to trigger dopamine loops content that includes violence, self-harm, dangerous behavior sexual or explicit content extreme emotional manipulation content targeted at minors in unhealthy ways etc
1
0
0
Positive Viral Moments (non-addictive) Examples: Inspirational stories Breakthroughs Science discoveries Historical storytelling
1
0
0
What we're training on now: -Cinematic founder stories & launches -Kurzgesagt-style explainers -Acts of kindness & community wins -High-craft creative work (animation, editing) -Positive mental-health motivation. Explicitly filtering out ragebait, fearporn, and compulsive hooks
1
0
0
i've seen/heard enough..Elon might be the one to the top
Let’s see if @Grok 5 can beat the best human team @LeagueOfLegends in 2026 with these important constraints: 1. Can only look at the monitor with a camera, seeing no more than what a person with 20/20 vision would see. 2. Reaction latency and click rate no faster than human.
0
0
1
Let’s see if @Grok 5 can beat the best human team @LeagueOfLegends in 2026 with these important constraints: 1. Can only look at the monitor with a camera, seeing no more than what a person with 20/20 vision would see. 2. Reaction latency and click rate no faster than human.
5K
6K
79K
VSA will be the first app powered by Jarvimind:
veoscriptai.vercel.app
No guardrails. Just virality.
We’re building the intelligence layer that actually understands why videos go viral. Patterns, emotions, pacing, hooks, structures behind every 10M–1B+ view clip. VSA is the first product on top of it: type any idea → get a ready-to-shoot script engineered to explode.
0
0
0
Be part in training the viral DNA: https://t.co/LVusjSNyev
We’re building Jarvimind: a post-transformer foundation model whose entire training distribution is 1 billion+ proven viral moments — nothing else. The result is the first intelligence that doesn’t just generate text; it engineers addiction and predicts exactly what stops scrolls
0
0
1