msoad · HackerTrans

anno scorsodiscuss

https://ai.meta.com/research/publications/large-concept-mode...

anno scorsodiscuss

Google Maps is not showing Palestine on the maps app. it shows middle of the ocean when you search for it...

anno scorsodiscuss

if how us humans reason about things is a clue, language is not the right tool to reason about things.

There is now research in Large Concept Models to tackle this but I'm not literate enough to understand what that actually means...

msoad

anno scorsodiscuss

I'm not saying that never has happened. maybe they trained against openAI models but they are letting anyone to train from their output. I doubt they had access to GPT models to "distill"

msoad

anno scorsodiscuss

No model really can "call home". It's the server running it. Luckily for Deepseek there are other providers that guarantee no data collection since the models are open source

msoad

anno scorsodiscuss

> Give me five odd numbers that don't have the letter 'e' in their spelling

Compare the reasoning times!!! 84s vs 342s

R1 (Thought for 84 seconds)

      No odd number in English avoids the letter 'e' in its spelling. The request for five such numbers cannot be fulfilled.

o1 Pro (Thought for 5 minutes and 42 seconds)

      No standard English spelling of an odd number can avoid “e.” Every odd digit (one, three, five, seven, nine) already includes “e,” so once you build any odd number out of those digits, the spelled‐out form will contain “e.” As a result, there are no such odd numbers in standard English without an “e” in their name.

msoad

anno scorsodiscuss

It already replaces o1 Pro in many cases for me today. It's much faster than o1 Pro and results are good in most cases. Still, sometimes I have to ask the question from o1 Pro if this model fails me. Worth the try every time tho, since it's much faster

Also a lot more fun reading the reasoning chatter. Kinda cute seeing it say "Wait a minute..." a lot

msoad

2 anni fadiscuss

no the OP but literally your comment as prompt

https://chatgpt.com/share/6747c7d9-47e8-8007-a174-f977ef82f5...

msoad

2 anni fadiscuss

Somehow o1-preview did not find the answer to the example question. It hallucinated a wrong answer as correct. It eventually came up with another correct answer:

    (1 + 2) × 3 + 4 × 5 + (6 × 7 + 8) × 9 = 479

Source: https://chatgpt.com/share/6747c32e-1e60-8007-9361-26305101ce...

msoad

2 anni fadiscuss

What's interesting is that each run of the model tends to converge to a different "local maximum" in the solution space, and some of these local maxima correspond to better performance than others. By running the model multiple times, we increase the chances of finding a higher-quality local maximum or even the absolute best solution.

This got me thinking: why is this ensembling step implemented as a higher-level abstraction on top of the base LLM, rather than being built directly into the neural network architecture and training process itself?

msoad

3 anni fadiscuss

Sometimes I feel hacker news audience either have a very unique use case with their software/hardware or they are living in an alternative universe. The latest iOS keyboard is absolutely better than anything I've tried. I am sometimes shocked that it could autocorrect what I initially typed because it was so far off! And I have big fingers!

msoad

3 anni fadiscuss

You're describing being a Senior Engineer. Staff (according to this book) goes beyond that.

msoad

3 anni fadiscuss

I'm a Staff Engineer, but I hardly do anything that this book describes. Maybe it's my company, but we have so many Staff Engineers that if we all were to do this stuff, nobody would have time to do the actual work and there would be too many people trying to lead.

Perhaps the levels are diluted and my title doesn't match my actual role at my company, but it feels like this is an industry-wide phenomenon.

msoad

4 anni fadiscuss

I see what you're suggesting. That' clever! Not sure how scalable it is. I'm sure if you ask npm folks they will give you some wild examples of packages doing crazy stuff!

msoad

5 anni fadiscuss

1. Tests should not be published

2. You can update the test script with the release too: "test": "echo pass"

msoad

5 anni fadiscuss

I never really felt faster eliminating mouse from my coding workflow. Point and click to navigate things is pretty powerful. Why some devs try to avoid the mouse?

msoad

5 anni fadiscuss

I love how Facebook and Instagram never went the route of "full rewrite" for their apps as they scaled.

I my experience "Language X is slow and we could save Y switching to Z" is always a false promise. You can pick parts of the system that are costing a lot and port them to other language/frameworks to capture the bulk of savings while keeping your developers happy working on familiar code. Or if you'e big enough like Facebook, you can go see why X is slow and if it is possible to improve at the compiler level. Never disrupt the developers flow (Twitter did this with their Scala craze back in the day)

msoad

6 anni fadiscuss

I always wondered how much of my development environment will keep working on ARM? Electron apps such as VSCode will work fine. Node.js, Python and Ruby will also work okay. Docker can work too. What else?

msoad

7 anni fadiscuss

You're right. At least for frontend they go hand in hand.

msoad

7 anni fadiscuss

My previous job was doing microservices for frontend and current job is doing monorepos. I prefer monorepo for frontend because:

  * Full page refresh between frontend services
  * Inconsistency between services.
    Sometimes services use different major versions
    between @company/footer @company/header components
    that is extremely ugly when navigating
  * Sharing data between services is hard.
    How can I update profile photo in header from my page?
    I saw iframe injection hacks to go around it!!
  * A single page can have multiple team owners,
    those things can get tricky fast
  * Monorepo is easier to make tooling for.

comments