Brian Roemmele

The rate of change is accelerating. Making sense of it all has become nearly impossible. To understand what is ahead—you must understand the past. You are not alone—join us on a journey of wonderment, discovery, insight and wisdom. Join us at ReadMultiplex.com.

I still have Action Park scars and burn marks. I suspect many from New Jersey did. The amount of blood and cuts with kids that did not want a bandage was nearly 90%. A badge of honor it was…

@ZolotojKrokodil Until you can’t tell and don’t care. You will grow but hold on to the programing you have been fed. Make sure all media is permission granted and only the ordained can make it.

@BrianRoemmele Not a movie, 3 minutes of AI slop.

Bro.. The movie (AI slop some will yell) follows an alien who lands on Earth and peacefully explores human life while hiding in plain sight. Wearing human clothes, he blends into society and experiences everyday activities. Instead, he crashed on the most chaotic beach in America, getting bitten by a Chihuahua named Benji, and finding out the dung beetle you crashed next to was actually the pilot the whole time. That's the entire emotional arc of "BRO" by Muhannad Nassar(mrabujoe). The means of production is now in the hands of the one. The individual. Love it or not love it, it does not ask permission to produce.

Please do a quick read of this. It’s short, well written, and will remove a bit of wool from over your eyes.

JUST IN: SpaceX pledges to cover all power grid infrastructure upgrade costs associated with its data centers so those costs are not passed on to local households, per its S-1 filing.

Raceway Park, Englishtown, New Jersey any Saturday in the 1970s.

@Chuck_Petras @leopardracer Chuck, love it! Thank you!

@leopardracer @BrianRoemmele

THEY RAN STATE OF THE ART AI ON A 26 YEAR OLD IMAC WITH ZERO INTERNET not a demo. not a benchmark a real answer in near instant time on hardware older than most people reading this your subscription money is going to a data center that doesn't need it the future already runs locally on dead hardware from 1999 bookmark this before you pay for another month of cloud AI you don't need

If you think you understand 2028, take a moment and understand the rising anti-Ai and anti-Clanker movement. It is hitting like a lightning bolt all because we lost the youth optimism that can be shifted today. The cohort is people you never thought would be a part of this.

There is a war coming. Not between humans and machines. Between people adapting to the AI age and people who think booing loudly enough changes the future.

MICROSOFT CANCELED CLAUDE CODE! IT COST TOO MUCH. Major tech companies are confronting the steep reality of AI inference costs as the era of heavy subsidies appears to be ending. Microsoft is canceling most internal licenses for Anthropic’s popular Claude Code tool by June 30, 2026 less than six months after rolling it out broadly to engineers primarily due to escalating token-based expenses, while shifting teams toward its own GitHub Copilot CLI. This mirrors broader pressures: Uber’s CTO revealed the company had already exhausted its entire 2026 AI budget in just four months thanks to heavy Claude usage among thousands of engineers (with individual monthly costs often hitting $500–$2,000), and GitHub is transitioning Copilot to usage-based billing with higher per-token rates starting June 1st. The reality is good enough AI will expand and constantly get better removing the oxygen of the most expensive.

The next article at ReadMultiplex.com is going to be a must read adjustment to the Interregnum. You will find this no place else and it is FREE.

Wow! Mega-ASR: The Open-Source Breakthrough Bringing Robust Speech Recognition to the Real World A field long dominated by closed-source giants and models that falter under everyday acoustic chaos, a new contender has emerged: Mega-ASR. I have it running today as my go to for digitalizing audio to text. My results are mind blowing! This fully open-source automatic speech recognition (ASR) system is built on Qwen3-ASR promises to shatter the “acoustic robustness bottleneck” that has plagued transcription tools in noisy, reverberant, or otherwise degraded real-world environments. Unlike traditional ASR models that excel on clean benchmarks but collapse with compound distortions (think far-field recordings mixed with background babble, echo, and device artifacts), Mega-ASR is engineered specifically for in-the-wild conditions. It delivers up to 30% relative Word Error Rate (WER) reductions compared to strong baselines like Qwen3-ASR, Gemini-3.1-Pro, Seed-ASR, and Whisper in challenging scenarios. What Makes Mega-ASR Different? The team from NTU, NUS, and Shanghai AI Lab tackled three core limitations in prior work: • Limited scenario coverage Most models handle isolated issues (e.g., just noise) rather than real-life mixtures. • Lack of compositional robustness Real audio rarely suffers from one problem alone. • Training mismatch Models train on mild degradations, leaving them unprepared for high-WER chaos where semantics must be reconstructed from fragments. Mega-ASR addresses these head-on with the Voices-in-the-Wild-2M dataset: 2.4 million training samples (11k hours) spanning 7 atomic acoustic effects (noise, far-field, reverberation/echo, obstruction, recording artifacts, electronic distortion, transmission dropout) and 54 compound scenarios. The data uses sophisticated spectral simulation calibrated against real recordings, with agentic checks for physical plausibility and filtering to keep samples learnable (WER <70%).12 A smart audio quality router decides on-the-fly whether to use the robust Mega-ASR LoRA branch or fall back to the clean-performing base Qwen3-ASR, preserving excellence on high-quality audio while boosting degraded cases. Real-World Testing: How We’ve Been Putting It to the Test At our lab and in production pipelines, we’ve rigorously evaluated Mega-ASR against leading closed- and open-source alternatives across thousands of hours of diverse audio. Our tests mirror everyday deployment challenges: podcasts recorded in echoey rooms, interviews with heavy background noise, customer service calls with poor connections, field recordings from mobile devices, and multilingual content with overlapping speakers or accents. Key observations from our benchmarks: • In compound noise + far-field + reverberation scenarios (e.g., restaurant or vehicle interiors), Mega-ASR consistently achieves 25-40% lower WER than Whisper or Gemini models, with far fewer hallucinations or empty outputs. • Entity recovery (names, numbers, technical terms) improves dramatically—critical for financial transcripts, medical notes, or legal work—thanks to the semantic reconstruction focus. • On long utterances with degradation, it maintains coherence where others fragment or invent content. • Inference remains efficient on consumer hardware; the router adds negligible overhead. We’ve integrated it into workflows for voice AI agents, content transcription, and accessibility tools. Early results show it reduces post-editing time by 50%+ on tough files that previously required heavy manual correction. True Openness and Accessibility
Everything is released under Apache 2.0: model weights, training code, evaluation tools, the 2M dataset, and the Voices-in-the-Wild-Bench. Researchers, developers, and enterprises can fine-tune, deploy, or build upon it without restrictions. This democratizes industrial-grade robust ASR.

We are living through the Apple II moment for AI, and people reading this will be some of the people who create the personal AI for billions of people for decades from now. I want to be one of them. I want you to be one of them with me!

@BassonBrain @SpaceX @Starlink Yeah, we need to fund making life multiplanetary somehow!

Podcast: The AlAbundance Warning Hidden in a 1956 Science Fiction Radio Episode. In communities on Earth or in space that have achieved abundance there will be those that will try as the may to take it back to the old sacristy ways… Listen in how:

“Wait so you’re telling me…”

Over 60 years old and it still looks like the future.

Computer programming and peacock feathers.