oogabooga13

Probably among many reasons for the switch to Gemini for their band aid AI until they get theirs were they want/need.

Youtube music hasn't failed me and their "beyond the beat" AI DJ/Music bits feature has been really solid.

Definitely a tragedy, I just think at some point the LLM needs to stop under any context (role play etc.) Personally, as a heavy Gemini user I went into the settings and have explicit instructions to not be sycophantic, to never 'fear' pushing back on me, tell the objective truth, etc) just to file down the default state of the model which can be a bit overeager to please or solve the next issue on every interaction. It could be too easy to walk away thinking I am the next Einstein and that seems to be something Google could stand to work on a bit.

Glad to see Branch Education represented here.

I was in engaged in a similar discussion with a co-worker where I was defending my view that I'd rather read the news than try to trust a brief video clip presented to me. I feel like I'm constantly keeping an eye out for excessive biases creeping into my 'trusted' sources. For a civilian like me, it's quite hard to grasp what lens the story is being presented to me as. Casually reading the news isn't a thing for me...it usually involves a little bit of research to ensure I'm not getting duped. The dead internet theory is important.

so it's valid when we say "the news practically writes itself" lol. This is so fascinating!

OP here!

Some context on why this exists and the decisions behind v1.0:

The Problem I'm a photographer, and my workflow was broken. I'd come back from a shoot with hundreds of RAW files and face two anxiety-inducing tasks: culling the duds and naming the keepers. I'm folder-first—file names matter because they follow the image everywhere: Affinity, Da Vinci, Apple ‘Motion’ layer stacks, client handoffs. A properly named file is searchable on any system without special software.

I wanted to point the computer at a source folder, point it at a destination, and have it handle everything in between. Locally. No internet. No uploading terabytes to someone else's servers.

The Journey [ Screenshots and more at https://oaklens.art/dev ] This started as a ~300 line CLI script on an M4 MacBook Air. After a few rounds of deep research (shoutout to Gemini for helping me break through some implementation walls), I had something that actually worked for my daily workflow. But I wanted to keep the low overhead of the terminal while making it more accessible. Enter the TUI—with two aesthetic modes: "Warez" (demoscene callbacks for those who appreciate that energy) and "Pro Mode" (clean HUD + stats for studio environments). F12 toggles between them. Fully open source (MIT).

Technical Decisions:

1. No Prompt Boxes: I didn't want to "chat with my photos." FIXXER treats the VLM as a headless reasoning engine. You press Auto, it applies logic—naming, culling, grouping—without you ever typing a prompt.

2. Native RAW Support: Most AI photo tools assume JPEGs. FIXXER works directly with RAW files (.RW2, .CR3, .NEF, .ARW, 40+ formats) via rawpy. We extract embedded thumbnails when available or do half-size demosaic in memory—no temp files, no export step. Straight from camera to AI pipeline.

3. Why Qwen2.5-VL: We tested Bakllava, Llava, Phi-3-Vision. Phi-3 failed hard on structured JSON outputs. Qwen was the only model consistent enough for production—good spatial awareness, reliable JSON, runs well on 24GB unified memory.

4. Graceful Degradation: Local-first means dependencies can fail. Semantic burst detection uses CLIP embeddings, falls back to imagehash. Quality culling uses BRISQUE (essential for not flagging bokeh as blur), falls back to Laplacian variance.

5. Hash Verification: Every file move is SHA256 verified with JSON sidecar audit trails. This eliminates the blind trust problem—you get cryptographic proof that your files arrived intact.

Flexible Workflows FIXXER is modular. The full Auto workflow chains burst detection → quality culling → AI naming → archive, but each feature works independently. Just want to group bursts? Run that alone. Just want quality tiers? Cull button. For the simplest use case, there's Easy Archive: point it at a folder of images, and it AI-names everything and sorts them into keyword-based folders. That's it.

AI Critique Mode: Beyond organization, FIXXER can analyze any image (RAWs included) and return structured creative feedback: composition score, lighting critique, color analysis, and actionable suggestions. It outputs JSON you can save alongside your files. This is v1—future versions will offer critique tiers based on depth and processing time. Configuration & Tuning The default thresholds (burst sensitivity, culling strictness) are tuned for my workflow, but everything is exposed in ~/.fixxer.conf. If the burst detection is too aggressive or the culling too lenient for your specific camera/lens combo, you can tweak the engine parameters directly to dial it in.

What's Next (v2) Dry run mode currently shows you exactly what will happen before any bits move. v2 will let you edit individual AI names in the preview before executing.

After a few rides in a Waymo I'd say I am confident they will tackle this the right way. However for companies with a track record like Tesla and their autonomous efforts idk what it means to trust "their" stack...

To be fair if you work in a glass palace you might think the world needs glass everywhere lol... : https://static1.squarespace.com/static/5e949a92e17d55230cd1d...

If it hasn't already been mentioned huge fan of newsboat paired with Lynx in the terminal. Travels easily and with lynx browser kinda brings me back to a more focused reading experience.

Not saying it's your exact use case, however, in the "saved info" section of gemini I have a prompt about the llm letting me *know what's on it's mind " along with some other details to where when I am just "chatting" it has brought up relevant books to our previous discussion / projects. Alongside local events (bay to breakers, roots game memorial day weekend, some single events, etc within that first" hello" of our conversations and brought some news to the foreground that was relevant to me, although I wouldn't necessarily seek out that info. It's been so handy to bring relevant info into my hands in an actionable amount of time. Plain Jane gemini didn't offer those amenities but I was able to build them out.

Agreed! I use Gemini and have found that I've been able to successfully shape the tone of the outputs -specifically away from the overly cheerful default by using the "saved info" section where you can basically act like a director for it.

Frontline PBS has an eye opener of a short doc about aircraft maintenance being farmed out to 3rd parties. Worth the 20 minute watch: https://youtu.be/sw0b020OFj4?si=mqfVRkco6rzgrVra

It's definitely why I finally bit the bullet and spent about a week or so to refine all of my "focus" status in iOS and have them automatically engage depending on geo-location or specific time of day.

After a week of observing what worked/didn't work for me I enjoy my phone more and get the notifications I want when I want.

Before setting up the focus groups I definitely was receiving too many useless notifications damn near every hour.

Of all the streaming options out there (Netflix, MAX, insert-your-service here) I pay for Youtube Premium. Has paid for itself many times over, especially with YT Music which is really good -it pulls from YT so I can get those rare songs/mixes that some individual just decided to upload from their personal collection. Also can download pretty much any video for offline vewiewing in the app.

The "free" movie selection is also really good (no ad's in premium). It's curated (read not endless fluff) and I spend less time thumbing through the damn menus (looking at you Netflix) and just watching stuff.

As an example YT Movies>Free just released James Cameron's Doc: Deepsea Challenge right after the Titan implosion. This type of realtime, zeitgeist curation happens all the time in their "free movie section" If you are starting from 0 in the submersible space great way to break the ice and start to grasp what that type of exploration entails. https://www.youtube.com/watch?v=ZZD_nbS1_II

Been with them since the Google Play Music days, just a happy customer.

The best anti-ageing cure that works now might be an optimized diet, exercise, and sleep routine. No protein injections required!

There are levels to everything. When I found myself losing close to 60lbs the weightloss in a sense was easy with a very basic plan. IMO I feel the mental component is not brought into the equation enough.

After the physical weight loss I still had to work through the mental issues with losing what felt like half of me (positive but still strange to witness in the mirror), becoming visible in spaces (think socially) I was typically invisible (despite being so large), and transitioning out of a year long weightloss mode to a life long sustainable "maintenance mode" to keep the weight at bay but not go crazy.

While the weight has fluctuated a bit since the intial weightloss over 10 years, the bulk 45ish lbs has not come back. One key are for me personally was wrapping my head around all the stuff surrounding my unhealthy habits, many not tied to food and working to shore up those areas in my life as well. Mind. Body. Soul.

Just wanted to throw that into the ring. Everyone has their own journey and the steps are really basic from a nutrition stand point. It can be a fascinating oppurtunity to learn about oneself through focused observation.

The assault on the Start Menu continues as planned...

CNBC had an interesting video about the Tesla semi's that Pepsi Co. is using. I say interesting because it just broadly touches on the project and in certain segments feels more like a marketing puff piece than journalism. Nonetheless still some stuff that can be extracted:

https://www.youtube.com/watch?v=l-BVM673pDs

Agreed! Since it's in beta par for the course IMO. That being said, in a certain context it's just as fast or faster because I don't have to dig into the webpage/manual cruft to get that nugget of information I need to keep moving! Hope access remains available to all....do we need a "Wikipedia" type of AI (few barriers to access) available to all?

I like trying out the Kagi Web + AI search beta right now. Not only does it give you chatgpt3 style answers, it cites sources. Going beyond verifying the informatation (which is very comforting) I can dig into the weblinks on my own. (no affiliation other than a satisfied customer) https://labs.kagi.com/ai/contextai

Also I don't like the NeevaAI pricing break down style. The annual plan which cost more upfront is broken down into a monthly amount instead of the what I will pay today at checkout price. It brings back negative memories of signing up for Adobe Cloud.

I've been a paid subscriber for awhile now and I'd describe it as the fastmail to gmail.

Although fastmail is quite a developed thing you can see where kagi is heading the longer they stay in the game.

It's up there in value with Youtube premium for me...I will pay a reasonable price for a life without ads given the option for a service I depend on.

Just me 02 cents.

Having a virtual movie theater has been the killer feature for me. I sit in on my comfy couch and stream movies from my computer into the headset. Awesome. For me it is better than buying a real TV for the purpose of movie watching.

Also VR Chat (also an app) has been awesome for connecting and talking with new people. Environments from orbiting high above the earth in space to a rooftop restauraunt in Paris make it fun to spend a few hours with new people.

As other have said, inside other VR application there is a lot of "neat" "potential" but hampered execution due to software, hardware, or both. I always tell people we've made some gigantic leaps but we're still in the early stages.

MacOS/iOS (user) after completing the recent update cycle on all devices "passwords" now supports 2FA using touch -ID.

I've been using Bitwarden for years but integrated 2FA support from Apple has moved me over. I really despised having to switch between apps (I don't use sms 2fa when I can).

This was a way to get through a level in Tom Clancy's Splinter Cell (the original game) 20 years ago.

https://youtu.be/lVNlggJECwc?t=507

It resonates with me working as a delivery driver [opinions are my own] (contracted, pretty much all Amazon drivers in the field) in SF that previously had a decades long office job. I was burnt out and needed an immediate change from spending years mostly inside.

From a physical and mental stand point, the work can be quite a lot. Amazon doesn't take it easy on you. However, after gaining some conditioning I enjoy being paid to spend time outside. It's confirmed that I need to take more serious look into fields that let me work outside.

To sustainably preform at my optimum, nutrition and recovery are essential. During "peak" (highest volume of pkgs moved) my  watch will record anywhere from 10 - 14 mi a day of walking with over a +1,000 calories burned.

Navigating streets with tourist, delivering on hills with significant grades, abnormal addresses and entrances (it's never quite as straightforward as one might think), numerous not up to code unique hazards, and what can be daily route changes - [one day delivering to skyscrapers in the Financial, the next to apartments in the Tenderloin) makes for a stimulating week of work.

It's not for everyone but can be rewarding while allowing a more focused split between work/life rather than where you're never really "off" from work.

I can agree with your general time scales on self driving. Just today in SF I watched firsthand a "Cruise" self driving car completely wig out when faced with a double parked car (pointed in the same direction as the "Cruise") and oncoming traffic on the opposite side of the street.

When it decided to make its manuever (on coming traffic briefly stopped to allow the car to drive around the double parked vehicle) the car made erratic micro turns and short hard braking action (pushing the nose of the car down) once it entered the (stopped) on coming traffic lane. The occupants were definitely thrashed around a bit.

The attempt was not pretty and definitely not even close to human level proficeny.

A typical manuever one has to make these days in the Bay with street parking eliminated in many busy restaurant / cafe corridors.

Love these tips!

Personally not a huge fan of the Zeiss wipes, seems to be too much solution on them.

The Target Brand...yes Target Brand wipes are great. Just enough solution to clean and not be abrasive and also evaporate without wet streaks on the lens. Also extremely cheap (the bottle version of the solution is nice as well.)

I'm suprised to see that some say Youtube's search is bad. I often find myself wishing that that I could get a youtube level of accuracy in my Google searches. I don't usually scroll to find the result I'm looking for. I tend to find the big view and small view videos with the same level ease.

*Also a premium subscriber. Youtube in many ways is a necessary part of this version of the internet experience and I detest having to sit through all the ads...if I have to pay $10/month seems fair to me. Just my 02 cents.

It’s crazy to me that WebOS is still the future all these years later. Love ios but it still doesn’t come close to webos in actual functionality. Who could forget notifications? Still the best way I have seen them handled in a mobile UI to date.