HNHacker News
TopNewBestAskShowJobs

tagyro

764 karma · joined March 18, 2019

move along, nothing to see here
submissionscomments
tagyro··on How is ChatGPT's behavior changing over time?
I read the paper and while I agree that maybe math is not GPT-4 strongest point, I have noticed the same degradation in quality.

And as I know there will be many commenters asking, here's my experience.

Four months ago I released a mobile app that wraps the "Open"AI API and allows a more private use. The app also includes 15 domains with over 150+ editable prompts, specifically crafted to help users get the most out of GPT.

I initially crafted these prompts in English but given I wanted to allow also non-English speakers to make use of them, I started to translate them in Italian and German.

I wrote a small script that took each English prompt and, with some more prompt-fu, translated them to these languages, using GPT-4.

As I'm fairly fluent in both Italian and German, I was able to verify the quality of these translations.

Well, around end of May, when I first noticed some weird answers from GPT, I ran my script again and (surprise!) the quality of the translations is visible inferior.

On a different note, I noticed that from one day to the other, any prompt would get an empty reply in the app, only to discover that "Open"AI has also made a subtle change in the json format of the API response that made broke the parser in the app.

Admittedly, 150 phrases is not a huge sample size and I could/should have used the default json parser.

I don't trust "Open"AI and will, as soon as it's feasible, change to a different model.

tagyro··on 2500 Job apps, 46 interviews, 1 offer
I read the post and the comments here and I honestly can't wrap my head around some of them (comments).

Being jobless takes a huge toll on the mental (and physical) health and instead of being supportive and empathic, we deride the struggle that is finding a job in this market.

In 2023 only, more than 200.000 people have lost their job (source http://layoffs.fyi).

Companies use ATS to discard CVs that don't perfectly align, ghost candidates, make bellow market-rate offers, post fake job ads ...and fire people to show strong quarters and make VCs happy.

And when someone has the courage to speak out, we attack them.

[somewhat "self-promotion": if anyone has a hard time getting past ATS, I built a small website to help align their resume to a specific job ad: https://rolematch.me . The service is free, no data is gathered or stored and the invite code is konami]

tagyro··on Giant telecom company that once almost bought Apple is on the brink of failure
I worked as a consultant for about 3 months back in 2012 for Telecom Italia.

I had a contract with an agency, that was “selling” me to another agency, that “sold” me to Telecom Italia. I was making about 2000 euro per month, but Telecom Italia was paying closer to 6000 for me.

My colleagues were some of the smartest people I ever worked with (I was actually working in Tilab, their R&D department) but it was very clear to me that the direction the company was going was not right.

tagyro··on SoundStorm: Efficient Parallel Audio Generation
just to be clear, one could mistake them for some (voice) actor reading a book (maybe) but even to my untrained ear they sound fake and artificial.

Am i missing something?

tagyro··on SoundStorm: Efficient Parallel Audio Generation
I've wasted (counting) about 300 seconds of my life listening to these audio files and they all sound and seem fake...
tagyro··on Internet Archive expands Flash support
I have many fond memories of Flash. It was far from perfect, but it really brought together programming and art, to a level that I don't see today.

Around 2005(?) I remember a Flash conference in Turin, Italy where I saw many great interactive CDs and games from artists from all around the world. Tokyoflash, Yugo Nakamura and many more creators that were building experiences which I have a hard time finding these days. In the latter years, there was FWA (Favourite Website Award) full of great flash websites (still remember GotMilk).

And then came Flex and AIR, that made building "rich internet applications" fun.

I left web development when Flash died and coming back to it now, after almost 20 years, and honestly, I don't think we're any better.

Adobe being Adobe, never got to fix the security and performance of the Flash Player (a dedicated surfer can still find malware disguised as Flash Player).

tagyro··on Calls to investigate tax sites for sharing financial information with Meta
> I think you were downvoted because your comment wasn't substantive and perceived to not be on topic.

Could very well be, that's why I prefaced it with "off-topic".

> If anyone cares about their data privacy I would expect HN users to be on the frontlines.

There's a saying/cliché/myth: expectations (can) lead to disappointment

tagyro··on Calls to investigate tax sites for sharing financial information with Meta
Can't really make out the tone of the question, but I'll try to answer anyway.

If by selling we refer strictly to "exchanged data for money", then no, based on the Senate report there is not direct evidence of that.

Based on the same report though, the tax sites had business agreements with Facebook to include the pixel and, again, even if they didn't receive any money per-se for that inclusion, they certainly benefited from it (via Lookalike Audience and generally, improved ad targeting).

So, does Facebook (and Google etc) sell your data? One can argue that technically they don't. Are they making a profit from having (and using) that data? I'll let you answer this question.

tagyro··on Calls to investigate tax sites for sharing financial information with Meta
Tax sites are selling data because there is demand for that data, in this case from Meta (or Facebook/Instagram/Threads/WhatsApp etc).

This is not the first time they (Meta et all) were caught with their hands in the cookie jar and still, over 100 million people thought it's a good idea to give them even more access to their information.

It's sad and disheartening to realise, again and again, that the "regular" user couldn't care less. And based on the downvotes I received, neither do the hn users. Besides some imaginary internet points, more importantly, I'm loosing faith we will ever open our eyes and see the cancer that are these data stealing corporations. </rant>

tagyro··on Calls to investigate tax sites for sharing financial information with Meta
A leopard can't change its spots.
tagyro··on Calls to investigate tax sites for sharing financial information with Meta
off-topic (or not): the newly launched Threads app has over 100 million users already
tagyro··on Show HN : Nova - GPT with composable prompts, workspaces, + LlamaIndex
I got to play with Nova a bit, a couple of thoughts:

- the quality of the answers is not better than gpt-3.5/4. Maybe I had too high expectations, but I didn't notice any improvements over the default answers from the "open"ai ones (for example, the "apple test" - write 10 sentences that end with the word "apple", got a 5/10 - not great)

- sign in with google ...I know this is the "easy" path for implementation, but I imagine some people (me included) really don't want to login with brother g

- the tone of Nova is, imo, too friendly. I know this is a LLM, don't need to pretend like it's my friend/counsellor

I hope my feedback doesn't come of as rude, looking forward to the next iteration!

off-topic: on your website, you still have a (c) notice with 2022 ;)

tagyro··on Ask HN: Any hardware startups here?
FYI your website crashed my browser ...and now it's not accessible anymore
tagyro··on Show HN: Open-source resume builder and parser
Really nice! I like the simplicity and resume building UI/UX.

Focusing on the ATS part, I'm building rolematch.me to allow candidates to better align their resumes with a specific job ad.

tagyro··on Run and create custom ChatGPT-like bots with OpenChat
I went ahead and installed it in a proxmox container, was fairly easy on x64 (arm support would be nice). One suggestion: it would be nice to have a short-term memory - a la ChatGPT. With the token limit at 4-8k for GPT-4, it would be nice to take advantage of that with both the "long-term memory" (vector store) but also a "short-term" one (as in, sending the previous questions/answers for context).
tagyro··on Generative AI learning path
The course is not free.

To finish the course and get the completion badge, one must also do the labs, which cost money.

Also, given it's coming from Google, one also pays with "data".

tagyro··on OpenAI's plans according to Sam Altman removed at OpenAI's request
In general, the quality of the replies seems to have been affected. More specifically:

- the context window seems to have been reduced: I only have access to the 4k model, but even when the prompts and replies are well under this limit, the model seems to loose context, as in, the replies have no connection to the prompt;

- hallucinations: while this has been a general issue with LLM's and even taking into account a reduction in the context window, GPT-4 seems to hallucinate a lot more now;

- not answering: prompts for which I have received answers before get a "my cut-off date ..." answer now.

I used GPT-4 mostly via the API and I saved the threads (history) of all prompts I've been sending, so it's fairly easy for me to "test" the before/after changes.

Some tests that are easy to run:

- the "apple" test: write 10 sentences that end with the word "apple" - I used to get 8/10 most of the time, sometimes even 10/10. Now it's more like 6-7/10.

- a friend suggested the "jwst" prompt: when was the jwst launched? Depending on the answer (sometimes it says it is scheduled to launch, sometimes it answers correctly), subsequently ask "what hour and minute"

- use a 1-2k prompt and analyse the answer - this shouldn't be an issue with a 4k limit, but sometimes it will hallucinate or respond with something totally unrelated

temperature: 0.2, top_p: 1, max_tokens: 4000, streaming: true

edit: I've noticed that speed has been improved, but that doesn't really help when the quality suffers; also, data point of 1, but when using it from Germany (Europe?), the quality suffers, when prompting via vpn from US, it seems to be slower but the quality is better (but again, take that with a big grain of salt)

tagyro··on ChatGPT app for iOS
Thank you again for your email and for bringing this to my attention!

And really (really!) happy it's fixed, especially since I still don't know what might have caused it (I haven't made any changes and there's been no app update since v1.1).

tagyro··on Don't let Reddit kill 3rd party apps
I'd like to offer a different perspective: let reddit kill 3rd party apps.

reddit has only gotten worse in the past years and even if we manage to somehow "coerce" them into supporting 3rd party apps, the core issues will remain and will re-appear under a different form.

We've had forums before and centralising *all* into one place never seemed like a good idea.

Let's simply stop supporting user-hostile companies.

(yes, I know my comment might be perceived as "simplistic" or "narrow", but we, the users, made reddit what it is today - we also have the power and duty to undo that)

tagyro··on OpenAI's plans according to Sam Altman removed at OpenAI's request
Reading between the lines of the original article:

1. there are plenty of - consumer - GPUs, but it wouldn't be advantageous for OpenAI to fine tune the models to run on them, hence the artificial limitation to only use GPUs with 80+ GB. We already see plenty of projects working on adapting these models for 16-24 GB GPUs or even running on CPUs - that is obviously a problem for OpenAI

2. GPT-4 has been nerfed recently - there are plenty of discussions on this and I've also noticed it in my extended usage of both ChatGPT Plus and the OpenAI API.

Also, as a developer (disclaimer) of a project on top of the OpenAI API, I've received multiple messages from users mentioning the issue. Even in the OpenAI discord there are multiple threads of developers complaining about the change to GPT-4.

3. Plugins do have PMF - in the OpenAI discord there are multiple threads about accessing plugins via the API. There are also many projects working specifically on this (AgentGPT, repository loader etc).

My personal experience: as a developer, you can develop plugins but you can't charge for usage. At the same time, plugins are available only to paying users of ChatGPT Plus ...

4. OpenAI will wait and see what 3rd party projects get traction and once they see public interest in a project, will build that project.

I'm willing to bet the ChatGPT iOS app was built in a couple of weeks (at most) and was a direct answer to the multitude of apps on the App Store.

5. OpenAI needs the cover of regulation. They developed their models without it but now regulation could help to prevent others from reaching their levels.

At the same time, any open source projects that gain enough public interest will be copied/absorbed.

This reads like a KGB press release.

tagyro··on Show HN: HN Resume to Jobs – AI Powered Job Matching Tailored to Your Resume
Obligatory disclaimer IANAL (I am not a lawyer)

Using Google Analytics might have some GDPR implications in Europe. A more GDPR-friendly alternative might be Plausible Analytics.

Also, given the amount of information contained in a CV, even without a name, address, contact info etc, it could make the process of identifying the actual person fairly easy, thus it might have some GDPR implications.

You could have a look at some Privacy Policies from recruiting agencies [0][1], though those are (in my opinion) fairly "abusive".

my 2 cents: given this is a side project, I would not store any data at all

[0]: https://www.greenhouse.com/de/privacy-policy

[1]: https://join.com/privacy-policy

tagyro··on Show HN: HN Resume to Jobs – AI Powered Job Matching Tailored to Your Resume
I'll second this.

I would like to give this a try (been job hunting for the past 6 months), but I'm also a bit paranoid in sharing a lot of data, especially with a service that has no privacy policy.

tagyro··on More malicious extensions in Chrome Web Store
This actually hits very close.

As the developer - disclosure - of AKME, an iOS app that uses the OpenAI API, in a fairly private manner, I also noticed the myriad of apps that use misleading descriptions (as in, advertising the use of GPT-4, when, in fact, they use GPT-3.5 - ask me how I know), or don't actually use OpenAI at all (again, ask me how I know), buy reviews etc and charge user exorbitantly priced subscriptions.

Part of the generated profits are then invested in Search Ads, which pushes them to the top and trick more users into downloading ...rinse and repeat.

I made sure my app has no tracking, users that have an OpenAI account can use their own API key (via a Bring-Your-Own-Key model) while those that don't have one can buy in-app tokens (if I may say, decently priced, taking into account the "Apple tax") while also investing a lot of work into crafting dedicated prompts to improve the quality of answers.

tagyro··on More malicious extensions in Chrome Web Store
I used Charles proxy (basically a "man-in-the-middle") to monitor the network requests and the data transmitted.

Also, some iOS apps support running on an Apple Silicon Mac (with M1/2) and, in a similar fashion, one can use various apps to block or inspect the network traffic.

Yes, it's a bit more work - hence my complaint about doing free work for a ~$3 trillion company - but I like to know what data the apps I'm running are sending home.

I know I'm a bit paranoid but hey, we all have issues, right ...right? :)

tagyro··on More malicious extensions in Chrome Web Store
Do an experiment: pick one category of apps, whatever category, and look at the top 10-20 apps in that category. Half of them will have misleading data disclosure (as in, they say they don't gather user data or don't track users, when in fact they do), fake reviews (easy to spot 5-star reviews with same/similar text), don't actually have the advertised functionality etc.

The same with search: do a keyword search on the App Store and see how many results actually match the query and how many are ads.

This is not just my opinion - it's fairly easy to find multiple security researchers documenting these cases.

I have spent countless hours (if not days) reporting apps to Apple - basically doing free work - when Apple touts their App Store security and review process.

Also, the parent comment I replied to mentioned the developer fees that make the App Store safer for users - that is what Apple marketing wants us to believe, the reality is very different though (and I say that with a heavy heart, as an iOS/macOS developer).

Apple could and should do better and, until they do, they have no right to pretend the App Store is safe.

tagyro··on More malicious extensions in Chrome Web Store
The App Store is full of scam, misleading, data stealing apps.

A couple of weeks ago I reported some apps to Apple and instead of investigating they asked me to provide screenshots, exact steps to reproduce etc.

After I pay $99 for the subscription and share 30% of my revenue, I'm also expected to provide free work for a ~$3 trillion company.

tagyro··on Tesla instructed employees to only communicate verbally about complaints
off-topic: so many tesla fanboys downvoting anything remotely negative of their favourite billionaire
tagyro··on That ChatGPT iPhone app has serious privacy issues you need to know about
I'm working on adding Siri intents - something like activating Siri and saying "Ask AKME ..." and AKME would return to answer to Siri, which would read it out loud - but I don't have an ETA yet.
tagyro··on That ChatGPT iPhone app has serious privacy issues you need to know about
> Very nice app

Thank you!

> do you plan on adding voice?

You can select the text in each message and, from the pop-up menu, tap on "Speak". If you mean for text input, that can also be accomplished via native iOS functionality.

I also tested Whisper, but imo it's not really usable on an iPhone (at least, not yet). Initial reports of the official ChatGPT app mention excessive battery usage and the device getting hot when using Whisper.

tagyro··on That ChatGPT iPhone app has serious privacy issues you need to know about
There is a subtle, but important, difference between ChatGPT and the OpenAI API in regards to data usage - https://openai.com/policies/api-data-usage-policies .

ChatGPT is free but users pay with their data. Using the OpenAI API on the other hand costs money, but users data (the prompts they send and the replies they receive) is not used for further model training.

Additionally, when using ChatGPT, one can select to disable chat history, which prevents using the data for model training - https://help.openai.com/en/articles/7730893-data-controls-fa...

OpenAI does add some (imo, unnecessary) barriers to this, but that's another story.

Disclaimer: I developed AKME - https://apps.apple.com/app/akme-ai-knowledge/id6446436196 specifically for users that want to use the OpenAI API, but would rather not have their data used to further train models.

Users can enter their own API key (this implies they have an OpenAI account with a payment method connected) and pay directly OpenAI, based on their usage. The app has no tracking, monitoring, data gathering etc and all prompts are sent directly to the OpenAI API in this case.

Users that don't have an OpenAI account (or can't create one) can alternatively buy in-app tokens, though, to cover for the "apple tax", those are a bit more expensive than the OpenAI tokens. In this scenario, the prompt is sent to a (very minimal) proxy that appends my API key to the request.

← PreviousPage 4 of 6Next →