akelly

HBM uses about twice as much DRAM silicon per GB due to all the space for interconnect

I've heard GenAI.mil still has Gemini 2.5 only.

Fake tracking numbers are a common occurrence in scams now. Somehow the scammers are getting access to a database of real time legitimate tracking numbers, they wait until there’s a shipment in their database going to the same city as the buyer, and then use that tracking number to claim that they shipped the package. Maybe they’re paying a real merchant for access to their shipping database? Or are UPS tracking numbers short enough to brute force?

Higher frequency means smaller and therefore lighter transformers, which is very important on aircraft. Nowadays a DC voltage supply would be better, but DC to DC voltage converters didn't exist when the 24V 400Hz standard was created.

For mains voltage we use 50-60 Hz because lower frequencies work better with very large AC generators in power plants and, and lower frequencies are more efficient to transmit long distances.

Apple doesn't want you to know this but they do their mechanical and electrical design on Windows PCs (using Siemens NX and OrCAD respectively)

Wow I'm shocked how cheap this is:

Prices, to keep this data anti public and affective only 5 copies will be sold

price per copy is 400$ [ negotiable ]

exclusive sale, only 1 person : 1500$

if you want to have the data + vulnerabilities we used to get into the database - 2000$

Right I intentionally left off Step 1 from that chart to simplify the explanation, since it didn't seem necessary. Is Step 1 just for creating the ChatGPT content blocker?

The way they went from GPT-3 to ChatGPT is really quite genius. My understanding is that it's something like this:

1. Start with GPT-3, which predicts the next word in some text and is trained on all the text on the internet

2. Take thousands of prompts, generate several responses for each of them, and have human reviewers rank the responses for each prompt from best to worst

3. The GPT model needs a massive amount of training data, it would be cost prohibitive to get enough human feedback to fine tune GPT manually. So you train another model, called the reward model, to predict how the humans will rate each response. Then you train the GPT model against the reward model millions of times

5. Feed a small percentage of the output from that training process back to the human reviewers to continue training the reward model, based on heuristics like reward model uncertainty which predict how helpful the human feedback will be towards improving the reward model

6. Release ChatGPT to the public, and use user feedback like response upvotes/downvotes to further optimize the reward model, while continuing to train ChatGPT against the reward model

https://openai.com/blog/chatgpt/

https://openai.com/blog/deep-reinforcement-learning-from-hum...

A friend's 18yo brother was motorcycling in the mountains with their father, crashed and broke his femur. Ambulance would have taken hours, they had helicopter rescue insurance, but the only helicopter company that operated there wouldn't take it. Got a $25k bill for the helicopter ride and negotiated down to $16k iirc.

Where did all the money go?

One theory is that the crypto quant funds figured out how to exploit the Alameda FTX market maker starting in 2020-2021 to take tons of money from Alameda. But FTX couldn't just turn off the Alameda market maker because most of the FTX trading volume, and therefore FTX revenue, was these crypto quant funds taking money from Alameda. So if they turned off the Alameda money spigot then their revenue would drop off a cliff and they wouldn't be able to raise more money from Sequoia or the UAE. And the value of FTT was tied to the trading volume and was a huge portion of their assets, so if volume fell they would be insolvent.

Basically they turned customer deposits into revenue at pennies on the dollar.

High impedance (low current flow) signals are more prone to electrical interference. This is particularly bad when you're running long wires across a factory with big electric machines running. And if your signals are low impedance (significant current flow), then you get voltage drop across your long wire, so you can't use voltage-based signaling.

The rumor I read was the entire M2 line was designed for N3 and then that got delayed so they backported to N5P.

Sorry I saw this on Twitter today and it didn't include it.

Can someone test whether you can do this style of prompt injection to discover what the rest of the prompt is?

This is a whole industry in China. There are buildings in Shenzhen with hundreds of vendors selling scavenged storage and memory chips, because those are highly standardized. And discontinued chips from the 80s and 90s are salvaged and sold on ebay and AliExpress so people can repair their old electronics. I've even seen broken laptops get their CPUs desoldered and turned into desktops with custom adapter boards. They're incredibly creative when there's profit to be made.

Wow you're right, GPSTest on my 2019 Oneplus 7 shows L5 GPS satellites, in addition to L1 GPS, GLONASS, Michibiki, and dual-band Galileo.

Making miniature electronics is really hard!

Tesla has been shipping custom 14nm inference chips in cars since 2019. https://en.wikichip.org/wiki/tesla_(car_company)/fsd_chip

With transparent lenses and no tracking, how are these better than the non-transparent video glasses that have been available for two decades and never caught on? The contrast ratio on any transparent display is terrible since the pixels can't block light, only add light. The darkest black levels you can get is however much tinting you apply to your glass.

Looks to me like Lenovo developed some display tech for AR and then couldn't get the tracking working and decided to ship a tech demo.

Today there is $20mm a day of ETH being printed and given to miners, most of which is sold immediately to pay their operating costs. After the merge there will be only $0.5mm a day being printed and given to stakers. Plus the money printed will go to people who are ETH holders, and are therefore less likely to sell.

Bigger chips have much worse yields, which is why phone chips get the latest nodes first. No way they launch 3nm with M2 Max. I was expecting 3nm A16 and 5nm M2 Max for release this fall and then the 3nm M3 progression next year. But they normally start iPhone HVM in July so A16 must be 5nm, there's no way Apple would delay the iPhone launch to November, it's too soon for M3, I have no idea what they're producing either.

TSMC and Samsung inflated their numbers a decade ago, Intel inflated their numbers last year to catch up, so now the nm numbers from all the fabs mean roughly the same thing.

Neither CPU company lowers prices unless they have competition from the other.

TSMC was founded as a Taiwan government initiative, the government still owns 6% and has a representative on the board of directors.

USB-A ports output 5V as long as the device is on. But with most USB-C ports, to save power no power is supplied until the device requests what voltage it wants. Requesting 5V is very simple, just two resistors which cost fractions of a cent, but somehow a lot of cheap devices forget to include them.

https://kalshi-public-docs.s3.amazonaws.com/regulatory/notic...

This is a huge blow to political discourse. PredictIt has been operating since 2014 with the approval of the CFTC as the largest election betting platform in the US. Today they received notice from the CFTC instructing them to cease operations by February 15th without clear justification for the decision.

One year ago, YC backed betting market Kalshi launched with CFTC approval. They employ a former commissioner of the CFTC as Chief Regulatory Officer, and two weeks ago applied for CFTC approval to have election markets. And half a year ago, the popular unapproved blockchain betting market Polymarket was fined and shut down in the US by the CFTC.

Further reading:

https://twitter.com/NathanpmYoung/status/1555387422510264321

https://www.cftc.gov/csl/22-08/download

https://astralcodexten.substack.com/p/the-passage-of-polymar...

You might make more impact finding the highest paying finance job and then donating most of your income to climate causes.

If that was his plan, why didn't he put an out in the contract?

I've been looking for a reader the last couple days and Feedly actually seems the best option. Most RSS readers don't allow you to save articles for later and put them in your feed. Pocket and Instapaper do this but require you to subscribe to RSS through IFTT or Zapier which I tried and seem very cumbersome. How is this not a native feature? Pocket and Instapaper allow you to subscribe to email newsletters, Feedly only offers this on their $8/mo plan.

I would subscribe to Feedly except their desktop Firefox and Chrome extensions for saving a page for later are both broken! On mobile it works except ~10% of articles it says it cannot parse. Just put a link in my feed then!

Anyone have another option that supports RSS, save-for-later, and email?

Edit: Wallabag does RSS and save-for-later but not email newsletters

Instapaper can send articles to your Kindle which is intriguing

Figured out how to configure Zapier, it's not too cumbersome, so much better than IFTT.