Thread

1. Oras · 1789820747 · focus · HN ↗

   I played around with Jev last night and did it for classification tasks that I used Gemini 2.5 flash lite with.

   It’s a bit faster and bit cheaper, but this is compared to LLM. The consistency was nice to see, BUT, as someone who trained NLP models prior to LLMs, it’s just BERT with more data. I can see why people would want ready made one shot classifier, and I can see the value of sending multiple classifier in one call, but I wouldn’t call it breakthrough. And I believe many labs will replicate it in no time and might have it as part of their harness.

   I see it as a wake up call for the tech community to go back to basics for most tasks instead of relying solely on generic LLMs.

   1. tchalla · 1789823087 · focus · HN ↗

      Anyone who has worked in ML for 10+ years would already know that the usage of LLMs for everything is lazy, wasteful and a high degree of marketing on it.

      1. iforgotmypasswo · 1789828729 · focus · HN ↗

         Anyone who has designed circuits will consider CPUs wasteful compared to ASICs. This new FPGA technology is just a less efficient ASIC.

         That’s roughly what I’m hearing.

         The fact that general purpose intelligent classifiers can be dynamically hacked together by an LLM in real time to allow them to build evolving labeled and understandable networks that perform substantially faster than the LLM, and can act as an intermediate sorting and organizing layer for caching context or handling simple tasks, and a complete layman like me can assemble a teachable layer of these in a few days from an inexpensive service…

         That’s wild!

         And then you can identify where an expert system needs a more specific ML technique for efficiency within this network that overlays the SOTA model. Or manually adjust the stored context in each secondary “neuron”. And paths forward can run programs or take actions at relative high speed.

         And you can share these with others and improve them as a group.

         You could insert this at the datacenters at scale with a local supervising expert to prune and encourage proper growth. You could identify specific gaps in capability that need more training, and patch over them temporarily.

         Then you train those corrections back into the general purpose model, or you identify highly efficient subsystems for specific purposes.

         And this is just one way to use it. High speed intelligent workflows can live in this. There’s a spot for a local LLM to learn on the fly.

         Maybe I’m way off base, but for the non-experts Jev seems extremely valuable.

         1. bigyabai · 1789838318 · focus · HN ↗

            I don't think either of you are wrong. The parent's assertion is that we've known this for almost a decade. BERT was highly usable for classification and sentiment analysis a whopping 9 years ago, despite being less than 0.5B parameters large. Similar-scale models like FLAN-T5 showed that it could be improved without substantially scaling up.

            Today, we're extremely spoiled by trillion parameter-scale models. Our conceptualization of vibe coding relies on wasteful tool-calling paradigms, the one-size-fits-all mentality of LLMs is part of the marketing blitz to make people buy more tokens. It's lazy on the part of frontier labs, but also wastes electricity, time and money.

            1. AIorNot · 1789840181 · focus · HN ↗

               Lol your argument is the same as programmers who complain about Javascript and internet browsers being the most common interface for all solutions on a computer

               You guys dont understand that the Lowest common denominator ALWAYS wins - its why excel is the linga franca for most companies

               LLMS and AI coding are the new javascript easy way to build amazing things and that trumps the tool specializers

               Years of Big Data and Data Engineers building fit for purpose ML pipelines expensively working in a shadowy corner of the company have been replaced by the PM vibe coding a tool to categorize his emails by relevance

               1. bigyabai · 1789841068 · focus · HN ↗

                  Again - you are right, but it still doesn't refute the grandparent's claim that today's AI is lazy, wasteful and marketing-driven. There is room to improve, and if US labs don't take the initiative then Chinese ones will.

                  1. mediaman · 1789862198 · focus · HN ↗

                     It is lazy and wasteful if you ignore the costs of specialized skills in doing it the "right way." If you stop looking at things in a narrow technical frame, and look at it as an organization, it's not wasteful. And lazy is a useless pejorative used against products that let people do things easily. Lazy is good. When you learn how to make products that allow people to be more lazy, you will become successful.

                     1. intended · 1789890667 · focus · HN ↗

                        What results though. The people seeing measurable improvements to their core work with LLMs are coders.

                        Everyone else is taking over intern level work from someone else’s team. They are reducing the friction costs of talking to someone else, for about a 30% productivity gain.

                        Firms are trying desperately to automate their white collar workers, and that is following the same trend as all other automation projects, and ML/deep learning efforts in history.

               2. joefourier · 1789843358 · focus · HN ↗

                  There's no need for black and white thinking. Javascript and the internet browser are the most common interface sure, but there's still room for specialised desktop software, especially those that require serious performance like anything to do with 3d graphics or real-time audio.

                  But also, frontier LLMs are enormously expensive and slow. Using Astra for things like simple text classification is not going to scale, and you're likely to end up in the same boat as those people who saw their Vercel bill shoot up to $96k/week when their site got traction, if not worse.

                  1. avinash147 · 1789975989 · focus · HN ↗

                     [dead]

               3. throwaway_2494 · 1789860957 · focus · HN ↗

                  A personal saying of mine: In computers the second best thing always wins.

Unofficial Hacker News client; not affiliated with Y Combinator.