0:07
AI: Alibaba open vs closed AI 'canary in the coalmine'. AI-RTZ #1143
I’ve discussed for a long while how China’s tech/AI companies large and small have taken the lead globally in open source AI models this AI Tech Wave. The lead was ceded by Meta’s global Llama models lead a couple of years ago, as it shifted its focus to more closed efforts like Muse Spark models released recently.
And China’s DeepSeek and other AI companies have pressed the pedal to whatever AI metal they could get their hands on, to stay close to performance vs the leading frontier AI models from Anthropic and OpenAI in particular.
0:39
Rich On Tech Episode 182 - July 11, 2026
Marissa Strang walks us through the little bits of tech that actually shave dollars off flights and hotels—price‑alert bots, route‑optimizing maps, and a handful of apps that stitch together itineraries without the usual spreadsheet headache. She shows how a simple calendar sync can flag cheap windows, and why pulling in local transit feeds early saves both time and surprise fees.
Then Christopher Kelly flips the script, pointing out the design tweaks that keep sites and devices from feeling like obstacles as we get older. Things like larger tap targets, contrast presets, and voice‑enabled shortcuts aren’t flashy, but they quietly make the whole experience feel more welcoming.
If you want to jump in, the line’s open at 1‑888‑742‑4101, and you can drop a question by email. The show rolls live on KFI Saturday mornings, with a rebroadcast on WOR Sunday evenings, plus the usual streaming spots on fire‑TV, Apple TV, Roku, and Samsung smart TVs.
1:30
This Week’s Top 3 SREF Codes Creators Will Want to Bookmark
The top‑ranked SREF this week feels like a neon‑lit cartoon cell from the ’60s—bold silhouettes, theatrical expressions, and saturated reds, oranges, blues, and purples clash in blocky shapes, while a grainy texture keeps it from looking too clean. It’s a sweet spot between retro TV animation, vintage comic covers, and modern flat illustration, giving characters a playful surface that hints at darker humor or mystery. Think music covers, indie animation, retro sci‑fi scenes, or any project that needs a punchy, personality‑rich vibe.
The runner‑up leans into a cold, occult‑tinged cyber‑fantasy. Its palette of muted greys, greens, pale blues, and blacks creates a solemn, almost ritualistic atmosphere, and the linework blends mechanical forms with organic, anatomical details. Small glows act like sacred signals, guiding the eye without breaking the desolate mood. It’s perfect for world‑building heavy concepts—mechanical deities, cyber‑temples, apocalyptic cityscapes, tabletop card art, or sci‑fi book covers that suggest a larger mythology beyond the frame.
The third favorite swaps spectacle for warmth, using loose ink lines, colored‑pencil textures, limited palettes, and generous negative space. Imperfect edges and slightly exaggerated proportions give the work a lived‑in, sketch‑book feel while keeping the composition tidy. This style shines on children’s books, educational illustrations, pet art, greeting cards, cozy packaging, and light‑hearted indie games—anywhere you want tenderness and a handcrafted charm without sacrificing polish.
2:51
Robot Dogs, Teslas, and Rescue Helicopters: The UN AI Summit Was a Lot
The summit’s demo floor turned into a tinkering lab: engineers streamed code live, tweaking a robot dog’s gait in real time while a Tesla‑derived AI tried to predict traffic patterns for a simulated city. The most striking moment was a rescue‑helicopter simulator that let participants feed sensor data directly into the flight‑control loop, watching the craft adjust its route on the fly. Those hands‑on sessions underscored a gap—while the hardware can be reprogrammed in minutes, the policy frameworks lag behind, still debating who owns the data and how liability is assigned. Organizers left us with a clear sense that the real work now is stitching those rapid‑iteration cycles into a governance loop that can actually keep pace.
3:32
AI-Powered Infinite Test Prep (Part 1 - DMV)
I've been thinking a lot about how AI can help people like my mom-in-law, who's a smart aerospace engineer but struggled with the DMV test because English isn't her native language. I realized that combining infinite test question generation and language translation could make a huge difference. So, I started working on PassMyTests.com, an AI-powered test practice tool that helps non-native English speakers study and learn concepts in their native language, practice test questions in English, and get explanations in both languages.
The MVP I built allows users to choose a language and US state, and then generates a realistic AI-generated question with four choices in English. After selecting an answer, users see an explanation for the correct answer in both English and their native language. I've also experimented with increasing randomness in the question generation, but I'm not sure if it's effective yet.
One of the challenges I'm facing is that only certain types of questions can be generated, and I couldn't create image-based questions or complex situational scenarios. However, I'm thinking about how to address these limitations, such as using a dataset of DMV handbooks to create factual questions or having two interacting agents to generate and answer test questions.
The good news is that my mom-in-law passed her DMV test after practicing on PassMyTests for a week, and she loved the feature of seeing English explanations alongside translations in her native language. She said it helped her understand nuances in the questions and answers. Next up, I'm going to tackle a bigger standardized test in Part 2 of this series. If you have feedback or requests for PassMyTests, feel free to DM me anytime.
5:01
AI-Powered Infinite Test Prep (Part 2 - SAT Math)
This is Part 2 of the series, AI-Powered Infinite Test Prep.
In Part 1, I shared my mom-in-law’s struggles with the DMV test in English.
This inspired me to build PassMyTests.com which generates infinite test practice questions while translating explanations into her native language.
I contemplated other standardized exams, similar in multiple-choice format, but haunting many millions more people:
The dreaded SAT!
In 2023 alone, 1.9 million students took the SAT.
I'm embarrassed to reveal how many hours I’ve spent studying for it.
I’d shove big SAT prep books into my backpack, heading off to Barnes & Noble.
I was one of those kids!!
I decided to tackle the SAT Math test first.
My MVP goal:
Generate infinite conceptually plausible and realistic SAT Math questions, while translating the answer explanations for non-native English speakers.
First, I repurposed my DMV prompt:
Your task is to act as a terminal for written test taking the SAT Math Test.
Produce a random 4-choice question.
The question should pulled be from a random set of 500 questions.
The question should be in English.
Finally, return the output in the valid JSON of the following format:
{
"body": "Body of the question",
"choices": ["Choice A", "Choice B", "Choice C", "Choice D"],
"answer": <index of the current choice e.g. A is 0, B is 1, etc.>,
"explanation": "Short explanation of the answer",
"translation": "${translation}"
}
Sabrina Ramonov @ sabrina.dev
Next, I made a new route in my MVP:
Here’s a sample question and explanation translated to Spanish:
Source: PassMyTests.com - SAT Math
This MVP didn’t take long, but now the hard part begins.
Next, I share several challenges I encountered.
Sometimes, ChatGPT fails to include the correct answer as an option:
Sabrina Ramonov @ sabrina.dev
In the above image, the correct answer is 600 square meters.
But this was not an option.
To fix this, I added a verification step to my prompt:
Before returning the output, solve the question.
Check that the correct answer is present in the list of 4 choices.
Sabrina Ramonov @ sabrina.dev
Much better.
Sample questions generated by ChatGPT-4:
Sabrina Ramonov @ sabrina.dev
Sabrina Ramonov @ sabrina.dev
However, even if the correct answer is in the answer list, sometimes the "answer” index in the output does not match the correct answer index.
Recall my prompt:
"answer":
<index of the current choice e.g. A is 0, B is 1, etc.>,
I haven’t improved this.
I’ll try assigning a random number between 0 and 3 to the right answer, then inserting it into the JSON output.
Nonetheless, initial results seem promising.
It’s been years since I studied for a test, but the questions looked plausible.
I was curious:
Would in-context learning improve it?
In-context learning is an effective prompt engineering technique where you include examples directly in your prompt, enabling ChatGPT to tackle new tasks without finetuning.
I added 4 real SAT Math questions to my prompt:
Your task is to act as a terminal for written test taking the SAT Math Test.
Produce a random 4-choice question.
The question should pulled be from a random set of 500 questions.
The question should be in English.
Before returning the output, solve the question.
Check that the correct answer is present in the list of 4 choices.
Finally, return the output in the valid JSON of the following format:
{
"body": "Body of the question",
"choices": ["Choice A", "Choice B", "Choice C", "Choice D"],
"answer": <'correct_index'>,
"explanation": "Answer explanation, show your work step by step",
"translation": "${translation}"
}
Below are the examples of the type of questions you should generate. Each example is delimited by triple quotes. Use these example to create various questions.
"""
Maria is staying at a hotel that charges $99.95 per night plus tax for a room. A tax of 8% is applied to the room rate, and an additional onetime untaxed fee of $5.00 is charged by the hotel. Which of the following represents Maria’s total charge, in dollars, for staying x nights?
A. (99.95 + 0.08x) + 5
B. 1.08(99.95x) + 5
C. 1.08(99.95x + 5)
D. 1.08(99.95 + 5)x
"""
"""
x^2 + y^2 = 153
y = −4x
If (x, y) is a solution to the system of equations above, what is the value of x^2?
A. -51
B. 3
C. 9
D. 144
"""
"""
The function f is defined by f(x) = 2x³ + 3x² + cx + 8, where c is a constant. In the xy-plane, the graph of f intersects the x-axis at the three points (−4, 0), (1/2, 0), and (p, 0). What is the value of c?
A. -18
B. -2
C. 2
D.10
"""
"""
A researcher wanted to know if there is an association between exercise and sleep for the population of 16-year-olds in the United States. She obtained survey responses from a random sample of 2000 United States 16-year-olds and found convincing evidence of a positive association between exercise and sleep. Which of the following conclusions is well supported by the data?
A. There is a positive association between exercise and sleep for 16-year-olds in the United States.
B.
9:09
Three Different Digital Counters To Remind Us How Good We Have It
The first build leans on a CD4026B decade counter IC, so it’s the biggest of the three, but it still uses only discrete parts and a 7‑segment display.
The second version swaps the IC for an AT28C64B EEPROM, which acts like a tiny lookup table: it translates the binary count into the segment pattern, shaving off a lot of wiring.
The third design crams everything into an ATtiny24A microcontroller, making it the smallest footprint and the simplest to program.
Together they show that even a “one‑button” counter hides a lot of hidden work—whether in chips, firmware, or the tools you need to get them running.
9:44
Phoronix Premium 2026 Summer Support Special Ends Tonight
So Phoronix has been around for 22 years, which is pretty wild. They've been doing daily news on open-source and Linux stuff, along with hardware reviews and benchmarks. Now, they have a premium subscription to help keep the site running, and they're having a summer sale that ends tonight. If you're into that sort of thing and haven't subscribed yet, now might be a good time to check it out. The sale is probably a good opportunity to support a site that's been around for so long and still putting out original content.
10:15
This amazing Raspberry Pi 5-powered device can visualise wireless signals around you
I’ve been tinkering with a little kit that runs on a Raspberry Pi 5 and suddenly the air around me turned into a moving canvas. Under the hood it’s just a tiny software‑defined radio hooked to the Pi’s USB‑3 port, feeding raw I/Q data into a lightweight signal‑processing stack that pulls out Wi‑Fi, Bluetooth and even some sub‑GHz chatter. The code then maps each packet’s strength and frequency onto a hue, and the Pi’s GPU paints it in real time on a 5‑inch HDMI screen. Because the Pi 5’s faster CPU and better thermal headroom, the visualisation stays smooth even when you walk through a crowded office. It’s a neat way to see the invisible traffic that normally just sits in the background.
10:55
Humanoid Robots Controlled By Surgeons Did World-First Operation On Live Pigs
I was amazed by how the robot’s brain stayed quiet while the surgeon did all the thinking. They took a Unitree G1—basically a five‑foot, 60‑pound humanoid that normally costs under $15 k—and fitted it with enough dexterity to hold laparoscopic tools. The surgeon, sitting miles away, guided the robot’s arms in real time, and the machine actually pulled two live pigs’ gallbladders out without a hitch.
What’s striking is the contrast with the big, pricey systems we usually hear about. The da Vinci platform can weigh a ton and run into the millions, while this little humanoid can be shipped in a box and set up on a modest clinic table. Its lightweight frame and lower price point mean a small hospital could, in theory, have a robotic assistant without reshaping the whole OR.
If the technique proves safe for people, the idea is that surgeons could bring that same remote precision to places that can’t afford a full‑scale robot. It’s less about replacing doctors and more about extending their hands into rooms that otherwise wouldn’t have that capability.
