Anyone who has spent time with parrots would realise that they can understand the meaning of speech without knowing what the words mean. Then somehow the meaning of word parts, and then you will find them making new words out of other words. Very clever indeed.
So stochastic parrots could indeed be a good description of LLMs. But I think that she meant it as a diminishing term (against the technology) which is pointless. Probably more of a reaction against SV tech bros than more nuanced interpretations.
You can structure the system to cause an LLM an answer structured questions, Y/N or from a structured list of options, or with numeric weights against labels, and then use classical parsing to enforce a compliant response submission, and you do that for random batches of comments, then have another LLM summarise the survey response. It is quite possible to extract sentiment and key topics.
I don't think it's useful to throw your hands in the air and say LLM context statistics are ungovernable. Both context hardening and action shielding architectures are addressing this, and in combination with well designed pipelines the risk is controllable.
It's important to note that this is in relation to real time position estimation. You can collect the signal measurements and process offline later for telemetry reconstruction.
You can still build a system that isn't vulnerable by limiting the API the LLM can access. A process consuming untrusted comments for summarisation shouldn't have access to account private data, it should just deliver a summary report. Another process can them scan that and remove/disable links etc.
Starlink is going to completely gut the local telcos, terrestrial fiber is never going to have the $ to achieve critical mass. So when, eventually, SpaceX or the US Govt wants to apply leverage over Africa, they will have to bend the knee. Or we'll have a hiatus in launches or a space war, and as they constellation burns up so will internet access. Africa will be in the dark.
Crime benefits from network effects even more than regular businesses, you have to attack them via every opportunity.
However you are putting words in my mouth, in a typical American style of prosecution-->banning. It is quite possible to legalise and regulate marijuana. American halfway legalisation creates an industry which funds OC and can't be prosecuted. Canadian legalisation creates a revenue source for the provincial governments while providing vertical integration and control, and the volume of illegal weed has plummeted.
A kilo of weed is clearly a dealer, and part of organised crime. The same people are deeply involved in forced sex work and people trafficking, extortion, illegal weapons, etc. There is a clear difference between end users and small time dealers and the distributors.
Some idiot drew the security boundary around all code delivered by VW, instead of around the car and it's supply chain. It's analogous to believing they could manage the road the car is driven on (something that Tesla seems to believe).
The liability culture at companies like VW is delusional, and they are deeply uncomfortable with software. They are suffer from individual departments and groups having individual authority and no drive to collaborate. Hence schizophrenic policies.
I don’t know who you are. I don’t know what you want. If you are looking for ransom I can tell you I don’t have money, but what I do have are a very particular set of skills. Skills I have acquired over a very long career. Skills that make me a nightmare for people like you.
By logging I mean a gateway/proxy which logs connection metadata, not client side logging. Possibly this is something that could be added to a private relay. With a network of private relays you could form admission controlled virtual overlay networks for each application.
For key revocation, it is necessary to be able to invalidate a server accepting a key, without necessarily having the key. E.g. maybe you have a pre-generated proof enabling revocation.
Observability is super important for organisations. Especially with LLM servers and MCP interfaces popping up everywhere.
This is for clients paying millions for deliverables, with high stakes deadlines. Expertise in the domain is in short supply. Data and model control is very important, so relying on AIaaS was already risky. You can LLM enable engineers without big AIaaS risks.
A 16C Ryzen 128GB with 96GB Blackwell is ~US$16k, quite reasonable for a worker billed at $300k. In fact so reasonable it's worth having AI enabled backend for lots of things.
I know some orgs that are already spending more on tokens than developer salary. That's what unlimited use leads to. Teams of agents running on expensive models, agents designing and running test suites with barely any oversight.
Programmers are used to paying nothing for tools. A basic laptop (SSD, multi core, 16GB of RAM) is hugely powerful if you are building in C/C++/Rust, even python. But all of a sudden it's no good, and we're back to using someone else's computer, hiring our tools every day. Worse, we get a different model every day, and maybe we aren't allowed to borrow the good tools some days because some mafioso are shaking down the manufacturer.
Most other trades need to invest significantly in tools. If you want good tooling, you really want 64GB of GPU memory (e.g. 2x 5090) and 96GB of RAM. If I'm paying $200k for an expert engineer then $50k every other year for tooling seems pretty reasonable.