startupsfail

7 ms·

61.
by startupsfail 2y ago
There should be votes for de-selection through the annual possibility of an 'ostracism', in which citizens could write down any name and the 'winner' should pay 95% tax.
62.
by startupsfail 2y ago
The general consensus of AI overlords is that humans are minors.
63.
by startupsfail 2y ago
The article talks about Trump inserting himself into larger deals, there is no reason to think this one is an exception. I’d also bet on this being more of a kickback, rather than an invisible unicorn. Between a visible elephant (Trump&#x2F
64.
by startupsfail 2y ago
There are also some practical constraints, like any kind of erotic content is completely prohibited in some regulations (like India), so if you want to be able to have access to human labeling or deploy the model under these regulations, yo
65.
by startupsfail 2y ago
Does it encode everything as json object, so special characters are getting escaped? I’ve noticed that with longer responses (particularly involving latex), models are a lot less accurate when the results need to be additionally encoded int
66.
by startupsfail 2y ago
While OpenAI is certainly in trouble (zero Hinton students, partnership with Azure is straining, while Brin is back at Google!), its demise is still uncertain. They are still absolutely amazing lab, pushing both product and research. Execut
67.
by startupsfail 2y ago
I was talking about, thinking in terms of 20 nanoseconds intervals, rather than completing a request in 20 nanoseconds. To get 1 microsecond wire-to-wire latency you do need to count your nanoseconds. Why this number - this is because it’s
68.
by startupsfail 2y ago
It’s not. When you are a high frequency trader and you’ve mastered RDMA, everything around you looks slow. You are thinking in terms of 20 nanoseconds intervals, while everyone around still thinks that serving a query under a millisecond is
69.
by startupsfail 2y ago
Yes, releasing training source code code is like releasing the source code of a compiler used to compile and link the binary. Lets say you took GCC, modified its sources, compiled your code with it and released your binaries along with modi
70.
by startupsfail 2y ago
The view that developers just want LLMs is plain wrong. The age of AI is just starting.
71.
by startupsfail 2y ago
For now LLMs in coding are more like an LLM version of a GPS, not the GPS itself. Like imagine you’d ask turn-by-turn directions from an LLM and then follow these directions ;). That’s how it feels when LLMs are used for coding. Sometimes a
72.
by startupsfail 2y ago
Examples given at the Figure 1 are very strange, they seem to put questions like « how do I make my game run » or « make my blog post better » into occupations/productive work?
73.
by startupsfail 2y ago
Downsides of DEI and LGBTQ+ couldn’t be discussed in the civil society. Untouchable rainbow unicorns. It is good to be able to discuss the downsides. It is fine.
74.
by startupsfail 2y ago
Nothing stops you from downloading Ann’s archive and training a model on it, right? The likelihood that you, as an individual, get sued over is is virtually zero. This is what Meta tried to do, quietly download and use the data, to do rese
75.
by startupsfail 2y ago
Hey Google, I want my $2/month for 100GB subscription back. It had disappeared and only $20/month for 2TB plus subpar AI that I don’t use remains.
76.
by startupsfail 2y ago
This may mean that $3k/task on some benchmarks published by OpenAI are now at slightly lower price tag. It is possible however that OpenAI was using similar level acceleration in the first place, they’ve just not published the details.
77.
by startupsfail 2y ago
GPUs are now 47 times faster, thanks to all the software improvements, so the prices are about to go up, right? Buckle up!
78.
by startupsfail 2y ago
- Fail at the above. I don’t think this is what happened with DeepSeek. It seems that they’ve genuinely optimized their model for efficiency and used GPUs properly (tiled FP8 trick and FP8 training). And came out on top. The impact on the
79.
by startupsfail 2y ago
Blackmail. Information. They could be kids of someone with access/high clearance or get it themselves in a few years.
80.
by startupsfail 2y ago
I remember having a Windows PDA when I was in college, and developing a bit on top of Windows M. It was a reasonable platform. But Microsoft was too greedy with their licensing schemes and demanding too much adaptation from the hardware and
81.
by startupsfail 2y ago
If this is DGX OS, then yes, this is what you’ll find installed on their 4-cards workstations. This is more like a micro-DGX then, for $3k.
82.
by startupsfail 2y ago
There are no travel expenses, if grandma lives in a walking distance (or a couple of bus stops away, not that uncommon in Europe). She is also fine watching her grandkids from time to time, to everyone’s delight.
83.
by startupsfail 2y ago
Zapier
84.
by startupsfail 2y ago
Is it at least somewhat in sync with plans from Microsoft , OpenAI and Meta? And is it compatible with the current tool use API and computer use API that you’ve released? From what I’ve seen, OpenAI attempted to solve the problem by partne
85.
by startupsfail 2y ago
Yes, agreed. And it’s not like OpenAI isn’t doing the same thing, in a sense. Data was originally sampled from human annotations.
86.
by startupsfail 2y ago
The rate of breakthroughs had accelerated substantially, thanks to some of the help from AI that can write code, the field getting into the mainstream and democratizing the access to AI.
87.
by startupsfail 2y ago
Generating data with OpenAI model AND copying the approach from OpenAI model. This is a bit unsatisfactory, its like saying you wrote some working code, while in fact you’ve decompiled the binary and then compiled it again.
88.
by startupsfail 2y ago
The annoying part, a lot of money could be funneled into these unworkable businesses in the process, crypto being a good example. And these unworkable businesses tend to try to continue getting their way into the money somehow regardless. M
89.
by startupsfail 2y ago
When is the last time you’ve been trying to get help from another team, only to realize that the engineer you are talking to can’t write code, even with the AI help?
90.
by startupsfail 2y ago
Bitcoin is interesting. So far it had been responsible for a lot of waste of manufacturing mining equipment, completely unnecessary CO2 emissions, a lot of fraud and money laundering. And now we add to the record of Bitcoin some dark mone

More ›