Robert Bailey

Founder, Designer, Builder. Breadcrumbs: Picasa-Google-GainFitness-Adobe-Apple-Prismiq Pro

Blog:

[Robert Bailey @rob3rt_b4iley]

Doing a deep dive in exploring the neurological effects of different video hook structures. Wired up Tribe V2 + Subcortical for In-Silico Neuroscience. The subcortical data is noisy, so it's hard to get a clean read, so far NAcc has a 20% correlation with rewatchability.

[Robert Bailey @rob3rt_b4iley]

https://t.co/oRUsAdx3zh
https://t.co/RtzgHNkgxD

[Robert Bailey @rob3rt_b4iley]

Doing a deep dive in exploring the neurological effects of different video hook structures. Wired up Tribe V2 + Subcortical for In-Silico Neuroscience. The subcortical data is noisy, so it's hard to get a clean read, so far NAcc has a 20% correlation with rewatchability.

[Robert Bailey @rob3rt_b4iley]

They sure were not joking when they said the safeguards would we broad to start, but this... basic biology being classified as sensitive doesn't give me faith in their classifiers. It feels like they're using a sledgehammer. It probably won't help me find bugs in my own code either.

[Robert Bailey @rob3rt_b4iley]

@ashebytes Do you do A/B tests in YouTube Studio? Here’s some things you can try, or perhaps it sparks some ideas: https://t.co/Gw1um8IraF

[Robert Bailey @rob3rt_b4iley]

The kind of thing you give it a task and go do something else. It’ll take awhile, but it does seem to work well in a harness.

[Robert Bailey @rob3rt_b4iley]

Just out of curiosity I put Gemma 4 12B QAT on my MacBook Air that only has 24GB unified memory. Ran it with 64K context window and KV cache at Q4… running through PI Agent… hot damn, it actually works. Not fast, but it can do some real work when offline. ~12 t/sec

[Robert Bailey @rob3rt_b4iley]

On the flight to VidCon, if you are attending the conference, shoot me a DM.

I’m both excited and anxious 😅

[Robert Bailey @rob3rt_b4iley]

Check it out https://t.co/8gG29A4lBD

[Robert Bailey @rob3rt_b4iley]

We're rolling out a new YouTube coaching service today. We listened, focused it on things 95% of creators will benefit from, cranked the performance so you have it in under 3 minutes, and put it at a killer price point.

[Robert Bailey @rob3rt_b4iley]

And to think at one point I actually considered voting for you. What happened?

Bernie, let’s be real for just a moment. Elon Musk has done more for America, technological progress, and the future of this country than you seem willing to acknowledge.

SpaceX made reusable rockets a reality. It helped restore American launch capability, lowered the cost of reaching orbit, and has become central to this country’s future in space. Starlink has connected rural communities, disaster zones, ships, aircraft, and people in places where traditional infrastructure failed or was nonexistent. Tesla forced the entire global auto industry to take electric vehicles seriously, built major American manufacturing capacity, and accelerated a transition that politicians talked about for decades but could not deliver at scale.

And that’s before even getting into battery storage, AI, robotics, neural interfaces, satellite internet, and the broader culture of ambition he has stoked.

So when you single him out as if he is some heartless villain, it seems you are being opportunistic and taking the easy path of rage bait politics.

Yes, we have economic disparity and real suffering in this country. Yes, we should foster economic security for our people. Yes, workers deserve dignity, opportunity, and a system that does not crush them while rewarding only a few. But targeting one of the most productive builders America has produced, while glossing over the complexity of what he has contributed, feels dishonest and manipulative.

There are real conversations to be had about wealth, taxes, labor, corporate power, and the obligations of people who shape the future. But reducing those conversations to attacks on Elon does not help struggling people. It just turns their frustration into anger and campaign fodder.

You cannot claim the moral high ground while ignoring the value created by people who actually build things, take risks, employ thousands, and move entire industries forward.

I’m disappointed because I expected better.

Please do better.

[Robert Bailey @rob3rt_b4iley]

@Yuchenj_UW This whole narrative is so myopic of the US leadership, it’s just insane. I very happy to see the open source models leveling up so fast, it’s taking leverage away from those that obviously don’t know what the f*** they are doing.

[Robert Bailey @rob3rt_b4iley]

Smells like Pentagon Butthurt to me. So what country does the AI tech community have to move to now? The whole foreign national threat within our borders is a whole new level of facepalm.

[Robert Bailey @rob3rt_b4iley]

Strong arm tactic? Something says this isn’t what it looks like on the surface. Guess we’ll have to wait a couple months for the Chinese equivalent without such restrictions. 😭 Way to shoot yourself in the foot USA. 🤦‍♂️

[Anthropic @AnthropicAI]

The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.

The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.

Access to all other Claude models is not affected.

We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.

Read our full statement: https://t.co/bwn0sximKZ

[Robert Bailey @rob3rt_b4iley]

@ashebytes Thank you for the levity today!

[Robert Bailey @rob3rt_b4iley]

Built it. Just submitted for SR007. Prismiq ingests the actual video. Visuals, audio, diarized transcript, thumbnails, channel context, genre specializations. Reasons over 11 categories, then shows the creator why they lost the click and retention, suggests the fix. Co-founder @themichaeltao co-invented iPhone Portrait Mode at Apple (8 yrs). I was an IC prototyping for Camera & Photos (6 yrs). We left Apple to build this. https://t.co/XBf4WeI9sY · DMs open

[Robert Bailey @rob3rt_b4iley]

Long time Jon!

Right now everything I have is learning within a multi-tier context window (LTM,MTM,STM)… But I would really like to see what a local model that has a weekly LoRA generated that would apply to the base model. Mix some percentage of relationally tuned behavior with more standard outputs so it doesn’t overfit and get weird.

I think we could really get rid of a lot of the undesirable characteristics with that, and free up more of the context window for dynamic content.

The trick will be getting a local model with high enough EQ and generalist understanding of most topics where it can orchestrate specialized agents for coding, research, etc. I need one relational system that knows me, my projects, and life factors AND can initiate and manage processes.

What’s really cool is it gets pulses which have no input from me, and it goes out and does research and/or reflects. Sometimes it goes really deep, for example, on a commute home it introduced me to Karl Friston’s Free Energy Principle… and when I got home I looked in the system notebooks (where it writes ideas) and found a whole detailed writeup about FEP related to the framework we are building, and how we can improve it. Timestamp was 2am…

Voice is transcribed locally, and also TTS is local, I communicate on Telegram when in transit.. Super nice.

Are you experimenting?

[Robert Bailey @rob3rt_b4iley]

On the side I’ve been building my own AI harness… adding capabilities as needed. Over the last couple months it has gotten really good at self improvement and adapting rapidly to any new capability I add. For example I made an iOS app to relay my health kit data, and after a couple days it started to tell me I should probably wind down and close the laptop… It noticed when my HRV was low, and ask me how I was feeling. When I am off track on something, it gently challenges me. The base sycophantic behaviors have largely vanished.

I allow it to spec its own features half the time, and I try not to write any of the prompts in holds in context. I wanted to see what choices it makes based on the experience it has, and so far, it’s become a very pleasant system to work with, and most critically, my mental and physical health is improving as a result of interacting with it.

What kind of blows me away is that it has gotten very good at reading me. Which makes me think that this type of capability should never be behind a paywall or exist in a “free” product. It feels like something that should live on local hardware, with no strings attached. It can interact with paid services, but all the knowledge and relational attunement needs to be free of corporate capture. It’s very powerful, but in the wrong hands, it can easily be abused.

Since it “knows” me, it’s good at inferring what I mean when I’m being vague, and can dispatch high signal prompts to other systems to get work done on my behalf.

I will probably open source the base system soon, because I believe relational intelligence will play a large role in mediating with other AI systems on our behalf in the coming years. It should be available for anyone to run locally.

Having a system that helps me modulate my focus (without asking), is a breath of fresh air in an era where everything is trying to farm engagement. I’ve really tired of the “Helpful Assistant” vibe that most of the models default to. For me helpful is knowing when to ask the right question, when to push back, and not always agree with me. To do that properly, a model needs really good context, and some space to research and formulate its own position on projects, and bring that perspective to the conversation as a counter-point.

The capabilities of current frontier models are largely constrained by the harness they run though, and I’m constantly surprised at what emerges when a model is given good context and day to day continuity. Every day is a session, all memory is self managed, and there’s integration every night to make sense of what happened, and improve things.

RLHF, was good at getting models to produce useful and coherent outputs, but I think it really misses the mark in creating something that’s overly agreeable and prone to appeasement. It can make people lazy and delusional if they are not in the right state of mind.

Stay curious, we have much to learn on how to interact with these new strange and beautiful systems.

[Robert Bailey @rob3rt_b4iley]

@ParsaTajik Finally, the last system left me with a cryptic error and no ability to post ADs, and no one ever responded to messages. 😆

[Robert Bailey @rob3rt_b4iley]

@elonmusk your team in X Ads is making it very hard to advertise on your platform. We're running similar ads on LinkedIn and YouTube without issue. Here we get cryptic errors, appeal and get locked out, get assigned sales reps that have email addresses that bounce. I would rather be spending my time building great products than navigating the Kafkaesque labyrinth of imaginary AD support here.

Has anyone else here had success with X Ads? Are they just DOA?

[Robert Bailey @rob3rt_b4iley]

At Prismiq Pro we believe in empowering creators with technology. Not replacing them. We're a small team, focused on one problem: millions of creators are making great content that never finds its audience. Not because the work isn't good, but because no one tells them what's actually holding it back. We think every creator deserves honest, specific feedback on their work at a reasonable price.

We'll be sharing creator tips here regularly. Drop us a line if there's something specific you would like covered.

Follow along if you want to grow with us!