antupis

12 ms·

151.
by antupis 3y ago
5 years ago it was hard to see , now I ampersonally using more chatgpt than Google.
152.
by antupis 3y ago
Is there currently companies where you can do this?
153.
by antupis 3y ago
I would not be surprised if this is the beginning of the end for the company.
154.
by antupis 3y ago
I like alignment more it is pretty quantifiable and sometimes it goes against 'safety' because Claude and Openai are censoring models.
155.
by antupis 3y ago
I bet Team Helen will jump slowly to Anthropic, there is no drama, and probably no mainstream news will report this but down-to-line OpenAI will shell off the former self and competitors will catch up.
156.
by antupis 3y ago
This is kind of true, I think programming even codellama or gpt3.5 is more than enough and gpt-4 is very nice but what is missing is good developer experience, and copy-pasting to the chat window is not that.
157.
by antupis 3y ago
I bet MS probably bankrolls a subsidiary or lightweight spinoff for AGI if they are under MS, they can keep the original research and code.
158.
by antupis 3y ago
GPT-3.5 and open source models need more RAG and prompt engineering but it can be as good as GPT4.
159.
by antupis 3y ago
Or Germans protect their biggest export.
160.
by antupis 3y ago
Google is living the innovator dilemma on steroids at the moment because Google Search + ads are so profitable everything else is a second thought and Google does not even bother to release stuff that they are developing.
161.
by antupis 3y ago
I think this tells more about company culture than the people themselves, some companies like small things, and others like cathedrals.
162.
by antupis 3y ago
Still that is rather small like here Finland average organization uses 34 – 57 million euros to lobbying. source https://julkaisut.valtioneuvosto.fi/bitstream/handle/10024/1...
163.
by antupis 3y ago
Personally, for me this flow works fine AI does the first version -> I heavily edit it & debug & write tests for it -> code does what I want -> I tell AI to refactor this -> tests pass and the ticket is done.
164.
by antupis 3y ago
Is the Apple Mx series about that?
165.
by antupis 3y ago
Rna vaccines might be that but it is still too early to tell.
166.
by antupis 3y ago
yup how I understand fine tuning is more about adding context and bigger picture. RAG is more about adding actual content. Good system probably needs both in long run.
167.
by antupis 3y ago
Who cares if it is open source model and weights are available. Nightmare scenario is that AI is behind some paywall and some entity can decide what goes in and what goes out.
168.
by antupis 3y ago
420.ai, 69.ai or 666.ai are much worse.
169.
by antupis 3y ago
100k tokens and good ide support would be great. Copy pasting back and forth with browser and IDE is kinda annoying and you always miss some context. I think model is now good enough but what is kinda missing is good developer experience eg
170.
by antupis 3y ago
Newton had horrible investments so his theories are bad.
171.
by antupis 3y ago
I think it is more about if you want go after general market you need to be bland that is why regular cars and clothes from HM look a same or biggest fast food chains menus look somewhat same.
172.
by antupis 3y ago
yup this is those cases where VC money makes sense, or at least tactic should complete different without VC-money.
173.
by antupis 3y ago
He has founded three successful companies Netscape, Opsware and A16Z so it’s not luck only.
174.
by antupis 3y ago
In Finland there is charging stations in almost every large grocery store and gas station.
175.
by antupis 3y ago
They kinda spinned their own genre 2004- most similar stuff what I kind know is shpongle(more downtempo) and Suomisaundi(own psy subgenre, weird stuff from Finnish artists)
176.
by antupis 3y ago
yeah I know but issue is not MBA it is what kinda culture you come and where you are good. Selling sugar water (you need to be like super good at marketing) and leading tech firm need different skills.
177.
by antupis 3y ago
Tim Cook is also MBA and it looks like it is working fine for Apple.
178.
by antupis 3y ago
Or generally what are good vector dbs have tried LlaMaindex, pinecone and milvus but all kinda sucked different way.
179.
by antupis 3y ago
Also this is search for LLM not for humans so optimal solution will be different. Or even with models it is not that hard to imagine that Mistral-8b will need different results than GPT4 which has 1.76 trillion parameters.
180.
by antupis 3y ago
Also work life balance is usually better in big corp you actually can clock that 9:5 and go home.

More ›