XR Multiverse

@GoogleResearch They have too much invested in expensive GPUs to ever introduce it at scale.

Introducing TurboQuant: Our new compression algorithm that reduces LLM key-value cache memory by at least 6x and delivers up to 8x speedup, all with zero accuracy loss, redefining AI efficiency. Read the blog to learn how it achieves these results: goo.gle/4bsq2qI

@iruletheworldmo People just forgot.

@iruletheworldmo Anthropic hasn't done anything with an .md file that custom instructions and GPTs didn't do in 2023.

i should probably make a prediction. anthropic will be the first lab to achieve agi/asi it’s fairly obvious that research and talent are the moat. now obviously you don’t get a seat at the poker table without a few gw’s and a private line with mr jensen. but meta and microsoft are proof that those things alone don’t count for shit. so ok fine, we’re in the era of research. so let’s look at who’s at the party rn. xai: still kinda stuck in the chatbot era, don’t feel as strong on agency and coding. huge re shuffle is a risk. could pay off. let’s see. google: the code red kinda worked, but not really. again model lacks agency, smart? yes. useful? i’m yet to see it. so who out of openai and claude seem to have the best research taste and shipping velocity? well, in the last eight months anthropic have been far in front. first to see how importing coding was, skills, computer use, mcps, claude code, co work. i could go on. they’ve even built clawdbot before the company that bought it…like, cmon sam. i’m an openai stan in truth. but. this is clear. and i wonder if it’s all powered by a) vastly stronger models b) vastly better research taste c) dario’s vision and focus big year i’d say.

@rohanpaul_ai It only took 3 years of using AI to remember all the things you tried to forget about software engineering that you learned in the last 10 years.

Harvard Business Review just published a piece. A good AI agent needs a job description, limits, and a manager. Because, AI agents can fail like employees with too much access and too little supervision. firms keep treating agents like normal software, even though the real risk is not bad text but bad actions. That changes 4 things: each agent needs its own identity and permissions, its own trusted data sources, hard rule checks between a model and any real transaction, and a full audit trail of what it read, decided, and did. So the safe rollout path is an autonomy ladder where agents start with drafts and recommendations, then move to guarded retrieval, then supervised actions, and only later get narrow bounded autonomy.

@GalvinAlmanza x.com/i/grok/share/5…

I don't get how people are planning to sidestep the very basic problem that if you don't have junior hires right now, you won't have experienced people 5 or 10 years later.

@AIContextWindow @mark_k The only capability they have is pattern matching. We manipulate the output of the patterns matched to suit our desired result.

@mark_k It’s very interesting that they do not understand their capabilities, but can just execute them.

At the heart of many misunderstandings about AI is the mistaken belief that AI models have introspection capabilities. In simple terms, you cannot simply ask them about their own capabilities. They do not actually know them and will often make something up.

@ujjwalscript Everyone should hire junior devs to fix issues using AI.

@ujjwalscript Tech debt costs: $0.70/million tokens. Tech debt no longer costs $150,000/yr plus overtime. Get over it.

The "10x AI Developer" is a MASSIVE lie. You are just a 1x Developer generating 10x the technical debt. The entire tech industry is high on the illusion of "vibe coding" right now. The popular consensus is that because Claude and Devin can spin up a backend in 45 seconds, software is now infinitely cheaper to build. Here is the provocative reality nobody is budgeting for: AI is about to make software engineering significantly MORE expensive. Everyone is cheering for code generation, but completely ignoring the Verification Tax. When an AI agent writes 5,000 lines of code, it is optimizing to pass the immediate test. It is not optimizing for human readability. It relies on brute-force loops, repetitive logic, and bizarre architectural shortcuts that just happen to compile. Fast forward 12 months. Your business needs to pivot, or a core dependency breaks. You are now staring at a 50,000-line black box that no human being actually wrote, understands, or can safely modify. You cannot simply "prompt" your way out of architectural collapse. When the machine-generated spaghetti finally breaks, you won't be saved by a $20/month LLM subscription. You will have to hire a top-tier Principal Engineer at absolute premium rates just to untangle the mess your "autonomous swarm" created. We are treating code generation as a pure productivity win, but code is a liability, not an asset. Stop measuring how fast your team can generate syntax. Start measuring how quickly they can debug it.

Everyone should hire junior devs to fix issues using AI.

@2sush You pay for an advantage to post on a site owned by an AI company, that scrapes your posts and trains AI with them. You stopped thinking for yourself when you joined to herd to destroy social media with AI.

The real risk with AI isn’t losing jobs. It’s slowly forgetting how to think for yourself.

@OmriBuilds Google hosts GPUs for Anthropic, OpenAI, Nvidia and Gemini. They can charge anything they want. The only winner is Google.

Who is winning the AI race? - Anthropic - OpenAI - Gemini

@wongmjane We'll be dead before that happens.

What’s the point of building a product, funding a startup, doing anything, or even existing, if AI will eventually be capable of doing everything?

@csaba_kissi It sounds like the team doesn't know how to use AI.

The scariest codebase isn't legacy code written by hand. It's fresh code generated by AI that nobody on the team actually understands. Keep in mind: documentation can't save you from ignorance at scale.

@Star_Knight12 You can ask Grok for the "most appropriate solution" to any problem in the world right now. Grok will consider all possible solutions and pick the 'most appropriate' one for the case. Try it. Solve something.

will AI able to solve all the problems of the world

@joni_vrbt @userluke_ Judges need to be wise and empathetic. AI is neither. It just sounds like it.

@userluke_ Do you really think so? Even if done right?

Shouldn’t court be run by artificial intelligence ??

@Yoshua_Bengio @ScienceBoard_UN If you think that's bad you should see the people it was trained on.

Evidence of deceptive behavior has already appeared in widely used AI systems, and the risk is expected to grow as AI becomes more capable, more autonomous, and more embedded in everyday decision-making. For further insights, see the latest @ScienceBoard_UN Brief to which I contributed. ⬇️

@dev_maims Are you complaining or bragging?

Hot take: Vibe coding only works well if you already know how to code.