Reubend

> A nicely aligned table with ANSI colors is for humans. An agent extracting a post ID needs JSON.

Wrong. While table formatting can confuse an LLM in some cases, a natural language output in pure text is almost always better than JSON for small amounts of data. After all, LLMs have more natural language training data than JSON training data.

The fallacy that LLMs need machine readable outputs just because they're machines is pervasive and it's a huge misconception about the way these models work.

On the other hand, I agree that large amounts of data should be outputted in a machine readable way so that the LLM can run scripts over it for more advanced parsing.

It can peruse research papers, world news, encyclopedias, product manuals, and dissertations.

> Yeah, I cannot imagine how anyone could learn anything well with access to AI.

You must not have much of an imagination then. Or maybe you're just being overdramatic? AI is arguably the best way to learn most subjects now. Frontier models have made a lot of progress on reducing hallucinations, and AI can teach you at whatever pace you're capable of learning it. There are very few topics it can't teach, and it can go into more depth than you'll find in any textbook.

I think it's supposed to be pronounced "Sandbox", but they realized that the word Sandbox would be unsearchable.

Wow, I love that realtime finance dashboard. It's a bold move to share that all publicly.

Sounds good, but my main question is: is this a fork, or a new backend they're building in (like MPS)?

It's a straightforward MIT license: https://github.com/calcom/cal.diy/blob/main/LICENSE

> IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

They'll sign a contract, and the contract will be very clear about whether using user prompts as training data is allowed or not. They're not going to care much about reputation; they'll care about the terms they sign with.

But the OSS license already absolves them of responsibility. This might just be to set the tone that security fixes won't be prioritized to the standard that they used to be.

I think you're right. Other providers can offer coding subscriptions that use in-house models, and this sets the stage for a Grok coding plan that's built in to Cursor.

$60 billion seems expensive, but it gives them a much better chance at competing in the market than if they started their own harness from scratch.

It's not just quantization. I verified that if you naïvely quantize to 1 bit from the original Qwen model (and set grouped scale factors based on what the original model's weights were like), it just spits out gibberish.

> One thought that suggests rearranging is not involved,a thought that does not require any knowledge at all: if it did involve rearranging, someone would certainly have added some order by scale factor tricks with linear interpolation by address offset to lose even less precision.

Can you elaborate?

Are you saying the original one worked with MPS? Or are you just saying it was always theoretically possible to build what OP posted?

After trying to understand their method, I think you're right. Doesn't seem like anything that I would personally call "diffusion". Much closer to MTP + speculative decoding.

Then again, their results with it are great. It would be interesting to benchmark it against standard SD on a model that already uses MTP.

> Also, any reason to imply "BridgeBench", apparently dedicated to AI benchmarking, wouldn't have run it more than once across the suite?

They didn't list a sample size of runs, didn't show any numbers for variance across runs, etc...

So while they may have done that behind the scenes and just not told us, this doesn't seem like a rigorous analysis to me. It seems to me like people just want to find data that support the conclusion they already decided on (which is that Opus got worse).

This page is buggy for me and doesn't show any plans. But when I go to their main pricing page, it's got some contradictory info about which plans include "Kimi Code". The $19 per month plan says that it comes with "Kimi Code available" but then shows an "X" near Kimi Code while the others have labels like "1x credits". So I guess they meant to say that it doesn't have it?

Because the website doesn't seem to show any sample size of runs, I assume they ran it once across the suite.

The models are nondeterministic, and therefore it's pretty normal for different runs to give different results.

I don't see this as evidence that Opus 4.6 has gotten worse.

Cool experiment! But the "CEO" agent picked the most boring possible items to sell: t-shirts and some bland art prints designed by AI. I would have loved to see more creativity given that they could have picked anything.

> I don’t see UBI anywhere on OpenAI’s roadmap.

Do you really think it's OpenAI's job to create UBI? Surely, if you feel that it's a good idea, then it should be the government who sets it up.

We can't just magically freeze the economy in time. If we conduct our industries inefficiently just to keep jobs around, we won't be competitive on a global market.

I have strong concerns around the inflationary effects of UBI, but whatever the solution there is, it's not the responsibility of private companies to organize their own welfare systems.

> AI has already stolen great amounts of money from a very large number of people all around the world, due to the huge increases in the prices of DRAM, SSDs and HDDs.

None of that is stealing. That's the free market.

AI is the best thing that happened to America in the last decade, and I dearly hope that politicians don't try to ruin it the way they're ruining other parts of the country.

I respect some of Bernie's positions, but his stance on tech is dangerous.

"Legibility" must be the wrong word because I can't understand what the author is talking about. Is he saying that the overuse of abstractions is ruining corporate culture? Or is he saying that the uniformity of corporate processes is becoming overbearing?

I think this needs to be re-written with different terminology.

"...you don't care about code review"

Code review is one of the things I care about most! In fact, now what we're in an age where many code changes are generated by LLMs, I think that code reviews are far more important than they used to be.

"Processes adopted by a company aren't about the end result of the work. Their stated goals and their actual goals are always distinct."

Wrong again. Often the goal is exactly as stated, for better or for worse. Let's take incident reviews as an example: their goal is to reduce the occurrence of emergency incidents in the future by learning from the mistakes that led to incidents in the past. There's no doublethink involved.

I would suggest that people stop overfocusing on benchmarks, and give this a try. Gemma 4 is performing really well for me, and seems to hallucinate much less than other models I tried in this size range.

This is pretty impressive. I think This sort of thing is a perfect fit for agentic coding because of the fact that you can compare the generated assembly afterwards as a safeguard/test. Plus, even if the code is messy, you can always ask the LLMs to do some cleanup passes afterwards.

You're absolutely right. And these Intel GPUs will also be much faster in terms of actual math than the M series GPUs that the Apple setup would have.

This looks great, but I'm wondering how effective this would be for full model weights rather than just the KV cache. Their paper only gives results for the KV cache use case, which strikes me as strange since the algos are claimed to be near optimal.

This is really cool research, but I'm wondering how much it slows down inference. The readme says that it's "...distinguished by zero overhead (no learned components, no entropy coding)" but does that really mean that this is a "free win"?

It's a very interesting concept, and I signed up to try it. However, after seeing the landing page, my first question was:

"Where's the data on accuracy?"

Backtesting is difficult to do correctly with LLMs, but because this is marketed as being for macro investing, I would expect to see a level of rigor and quantitative analysis consistent with that.

The Monte Carlo simulation engine sounds really cool, but is there evidence to indicate that it generates superior results to expert predictions, or to LLMs alone?

I actually think it would be totally fine for your beta version to have low accuracy numbers. After all, this seems to be something in the very early stages. But to have no quantitative analysis of your system's performance definitely makes me uneasy to trust it.

Seems like it does quite well on that particular benchmark?

A mobile failover would be cheaper and would give you better connectivity in heavy rain.

A 4G dongle can be purchased for $15, rather than $200 for a Starlink Mini. Then, let's say your main internet source fails and you need to actually use the backup plan beyond the standby amount of 0.5 Mbps. That will cost you a minimum of $50 for Starlink, versus roughly $25 for a month of unlimited cell service. As for standby costs, you can find phone plans for $5 per month tat give a small amount of fast data, as opposed to Starlink's unlimited amount of slow data.

But of course this only works for areas that actually have cell service.

Very interesting! Thank you for explaining this.