It's only good for things that don't have to be right. If I ask the same question to 3 different AIs, I get 3 different answers. If I ask the same AI the same question 2x I get 2x different answers. If I as technical questions (most dynamics and fluid mech) I get answers that are about 50% correct and the rest useless...So far AI is fine for search and simple questions, but nothing magical...
-
Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
Useful Applications for AI
- Thread starter amirm
- Start date
Check out the handiwork by @pierre, the gentleman who also brought us the marvellous spinorama.org. His SotF (Sound of the Future) project is about 1 million lines of code now. One great guy assisted by AI.
AutoEQ for rooms and headphones
Release is available at https://github.com/pierreaubert/sotf/releases/ Currently available: - MacOS (arm): native app, tui app, systemwide - MacOS (x86): built but not tested - Linux (x86): appimage format (and tar.gz but untested) - Linux (arm): currently building - Windows (x86): .exe...
thanks @NTK. I am using AI extensively for a few things and it is improving month after month:
- software design, architecture, security, ui and of course coding
- research (they are extremely effective at summarising areas of research and proposing new ideas)
- optimisation (see evo-hq for example on github: you have a goal, something you want to improve on. The model will try various strategy compute a score, pick the best one, iterate, etc) : it does work very well and allow to optimise things where you dont necessarily have experience with the field.
- mathematics and finding proofs in lean or quint
and more mundane stuff:
- sorting the computer
- tagging music
- music discovery
- mixing and mastering music
the list is endless, i have many ideas that are now doable, i need to get more focus to finish some of them.
i have a different experience on similar tech questions, codex+claude help me to build an excellent BEM and FEM software from scratch that works on everything from 1 cpu to a cluster of gpu. Physics currently supported are acoustics, aeroacoustics, beam, composites, compressible_flow, contact, eddy_current, elasticity, electroacoustics, electrochemistry, fracture, lumped_circuit, magnetostatics, maxwell, nonlinear_solid, piezoelectric, plasticity, porous_flow, semiconductor, shallow_water, shell, stokes, thermal, thermoelasticity, transport, and some turbulence. I have an optimisation software that can do CAD -> mesher -> bem+fem with shape and topology optimisation.
Is it working easily and is it doing everything for me, no.
if you want to try something fun: pick something to optimise and ask evo-hq to find better solutions. That did work great for ex with fft, blas or lapack like stuff, assembly. And since the tool never tired, you can optimise for neon, then avx then rocm then cuda etc
babayaga
Member
I use it in a few ways. I'm reading a book and there are several historical and biblical references that I was not familiar with. So as I'm reading, when I reach these parts, I pause, look up and explain to chatgpt what chapter I'm reading, it responds with the proper info.
In addition I like to mess with apple-script and python. When I run into an issue, I post my code and it tells me what my error is, explains it and provides a correct code.
When I am called to write anything, from a legally binding contract (yes, I am admitted in the applicable jurisdictions) to a Linux utility (for my own use only - I do not post anything to Github) to a friendly text or email, I find that Gemini is an exceptionally fertile brainstormer.
I also recently was able to use Gemini to backtest investment strategies more rapidly than I could have with any available tools five years ago.
I also recently was able to use Gemini to backtest investment strategies more rapidly than I could have with any available tools five years ago.
coding seems to be where it's most useful, my questions tend to be on mathematics. The problem is that if it gives a wrong answer, there is no procedural logic you can go in and fix, you have to update the training set and hope for the best. Additionally there is not procedural logic you can audit, it's good for certain things, but no one is automating the running of a nuke using AI
I discussed a good outcome I had with AI in this post
www.audiosciencereview.com
Master AI (Artificial Intelligence) Discussion/News Thread
I have tried many search engines and Perplexity has proven most accurate and reliable in my testing. YMMV! As I have gained experience with Perplexity, I have refined my queries. Most importantly, I have instructed it to use only original source documents - not to use news reports or other...
www.audiosciencereview.com
kemmler3D
Master Contributor
My employer is a software company and we're "all in" on AI. Even though I am not an engineer, I've found it (mostly Claude) really useful for searching internal documentation, and ESPECIALLY for doing database queries and data analysis. Stuff that would have taken me days (if I could do it at all) now takes minutes. It's also dynamite for list enrichment although I don't do much of that.
I've also been able to create a really fancy revenue projection app by feeding in some known values for our business and asking it to create a month-by-month model, with a level of detail / number of variables that would be maybe possible but very burdensome to do in a spreadsheet. Being able to do projections using sliders and such, rather than pecking at a spreadsheet, is really nice.
It's decent but not perfect at analyzing interview transcripts and similar. Can save time but occasionally misconstrues things people say, still.
Despite it all, I am acutely aware that I'm not doing "software engineering". I am throwing code in a box until it works. That's maybe coding, but it's not engineering any more than bolting two random drivers to a baffle is audio engineering.
I've also been able to create a really fancy revenue projection app by feeding in some known values for our business and asking it to create a month-by-month model, with a level of detail / number of variables that would be maybe possible but very burdensome to do in a spreadsheet. Being able to do projections using sliders and such, rather than pecking at a spreadsheet, is really nice.
It's decent but not perfect at analyzing interview transcripts and similar. Can save time but occasionally misconstrues things people say, still.
Despite it all, I am acutely aware that I'm not doing "software engineering". I am throwing code in a box until it works. That's maybe coding, but it's not engineering any more than bolting two random drivers to a baffle is audio engineering.
kemmler3D
Master Contributor
I used to agree with this more strongly, still do mostly. However, for things that can be done with code (SQL & python especially) Claude gets things right very often. In those situations it does a good job checking its own output against what you asked for, so when it gets things wrong it tends to correct them itself. Since it's running deterministic code to generate the answer, the answer tends to be correct and not loosey-goosey like pure chat answers.
I would say even from my POV you do need to be an engineer to get it to engineer code properly. The code will work but no guarantee it works the way you want it to. I would even catch it in certain technical mistakes, knowing next to nothing about the tech myself.
Skeptical_Aus
Member
- Joined
- Apr 25, 2024
- Messages
- 35
- Likes
- 77
I find gpt 5 really useful in the kitchen. I can tell it what I'd like to make and what ingredients I have and it produces a recipe to achieve it. Any steps not clear I ask for clarification, eg step 5 is a 10 minute simmer, is that lid on or off?
This week has been a week of spectacular Japanese ramen soups.
This week has been a week of spectacular Japanese ramen soups.
I would say that GPT / Gemini etc are generalised models and not capable of being right all the time (or even most of time on many topics), but you mentioned training sets, so I guess you already know that and are training or fine tuning your own models?
My general advice to someone who needs a software based solution to a problem and they need to know it will always give the correct answer would be "don't use AI". However, there are certain model types that can incorporate rigid rules, such as concept bottleneck models (CBMs). This type of model is different to retrieval augmented generation (RAG) where a user embeds their own ground truth or knowledge, typically in the form of vectors or graph structures, which is proven to significantly improve accuracy of LLM outputs. The difference with CBMs is that the model cannot choose to ignore the symbolic logic layer, whereas with RAG if the required information is not present in the vector or graph database then the model is likely to hallucinate something to meet the goals it was set during training.
CBMs are not a silver bullet as they comes with their own issues, such as usually needing the expertise to build the conceptual logic in the first place, and falling over if the concept is incomplete or inaccurate, hence being a bottleneck. Still, if accuracy and explainability are the top priority, and the problem cannot be solved without the use of AI, then CBMs are one of the more promising avenues worth exploring.
There has been plenty of work since 2020 on CBMs, but here is the foundational paper:
https://proceedings.mlr.press/v119/koh20a.html
OP
- Thread Starter
- #35
No way. IMDB list of actors for space:1999 is dozens and dozens of pages with hundreds of listings. Only the key talent has pictures. The rest is all names. If I had to do this manually would take me hours and hours.
Even with top talent it is more work to go to IMDB than to just ask Gemini in context of youtube page who is who.
I don't know the name of the reality show but I doubt that it has all the random customs agents in there.
Martin
Major Contributor
I design contact centers for a living and for the last couple of years the majority of RFI/RFQ have been for AI assisted human agent staffed contact centers. In the last month or two we've started receiving RFI/RFQ for full AI contact centers with no human staffing. From my own experience both as a designer and user we aren't ready to take that step.
Martin
Last edited:
Blew
Senior Member
You can work around this with LLMs by giving explicit instructions on how to answer. I found Gemini was often hallucinating on audio science related topics (much more so than ChatGPT) which my below instructions in my "audio" Notebook in Gemini help mitigate:
Answer the questions using verifiable facts. If the information is not explicitly known or if you lack sufficient data, state 'I do not know' or 'Information not available' then provide an educated estimate along with the probability of being correct. Be concise and use bullet points where possible.
So why did they take that 'step if there were ready for it ?
I'm glad we have found a suitable application for this technology after the hundreds of billions invested!
- Joined
- Aug 3, 2019
- Messages
- 1,799
- Likes
- 2,846
I use an AI mostly as a time-saving search engine enhancement rather than an interactive partner, e.g. when my e-mail provider recently nuked my POP3 access with what I assume was a mail server upgrade, I was able to retrieve Mozilla Thunderbird's encrypted local copy of my long-forgotten password with a free utility that the AI located for me, which in turn allowed me to get into my account on the provider's website and reenable POP3. I probably would have eventually found that solution without AI help, but the AI made the process quick and kindergarten simple. On the other hand, there have been numerous occasions when I wound up correcting the AI, which then managed a pretty decent approximation of a human mea culpa and corrected the answer for future querents. 
Similar threads
