jsemrau

20 ms·

271.
by jsemrau 2y ago
This shouldn't be downvoted as it raises an important point about privacy and safety.
272.
by jsemrau 2y ago
Same request as for the introduction of GPTs. Make sure you understand which part of your data you are giving the rights away. "Our use of Materials. We may use Materials to provide, maintain, and improve the Services and to develop ot
273.
by jsemrau 2y ago
Singapore also has great universities. Research into this area fits Singapore well.
274.
by jsemrau 2y ago
My feeling is the more public platforms are occupied by engagement economics, bots and fake everything, we humans meet in messaging groups like Telegram, Slack, etc that are not public.
275.
by jsemrau 2y ago
I hate when it outlook it shows me email following a keyword in "relevant" order and not in "recent" order.
276.
by jsemrau 2y ago
When I gave "Matt", my loyal local assistant[1], a voice xTTSv2 performed better for long form text. While in longform emotions seemed well balanced in the text, in short replies the emotion patterns frequently felt off and theref
277.
by jsemrau 2y ago
LCEL is such a weird paradigm that I never got the hang of. Why | use | pipes?
278.
by jsemrau 2y ago
Transformers != Modern LLMs. It is 100% an important invention, but other important tech like hidden markov chains existed long before.
279.
by jsemrau 2y ago
"The Overviews launch didn’t go well." understatement of the year.
280.
by jsemrau 2y ago
That's my Heygen/D-ID story to the T.
281.
by jsemrau 2y ago
"We are shifting our focus on GPTs to Commercial and Enterprise scenarios and are stopping GPT efforts in consumer Copilot" https://support.microsoft.com/en-us/topic/gpt-builder-is-bei...
282.
by jsemrau 2y ago
Interesting. I got quite good results for my longform substack by combining xTTS2 with Nvidia's Nemo.
283.
by jsemrau 2y ago
That's a whole category that should have been an app. Little throwback. In 2015, I built an audio-augmented reality running game using Bluetooth beacons. It worked better as an app.
284.
by jsemrau 2y ago
“The lack of processes and procedures was kind of mind-blowing,” sigh
285.
by jsemrau 2y ago
“nobody asked the question” That's such an odd thing to say for a search engine operator.
286.
by jsemrau 2y ago
They summarizer model has access to their core web page ranker. Since we saw Reddit and Onion posts in their result set for the summarizer, I'd argue that the goals of the core web page ranker don't align well with the goals of th
287.
by jsemrau 2y ago
"We hold ourselves to a high standard, as do our users, so we expect and appreciate the feedback, and take it seriously." In terms of AI Safety, I believe Anthrophic has built the right culture and mindset. In my opinion there is
288.
by jsemrau 2y ago
It's called governance. That's literally what a board of directors is for.
289.
by jsemrau 2y ago
I'm a lumberjack and I'm OK I sleep all night and I work all day
290.
by jsemrau 2y ago
There is some Monty Python joke about shrubbery somewhere in there.
291.
by jsemrau 2y ago
Haha, that resonates. When I built my LlamaIndex agent, I did same.
292.
by jsemrau 2y ago
I just hope the SEC enforces ISO 42001 violations. Especially for companies like Google.
293.
by jsemrau 2y ago
This is statistics though. Edge cases are nothing new and risk management concepts have evolved around fat tails and anomalies for decades. Therefore the statement is as naive as writing a trading agent that is 100% correct. In my opinion,
294.
by jsemrau 2y ago
> If you provide the tools, the will build Only makes sense if the incentive of learning that tech provides expectations of future benefits.
295.
by jsemrau 2y ago
I wrote a Langchain ReAct agent that searches the web for me using Wikipedia, Serpapi (google), and DDG. Works better than traditional search.
296.
by jsemrau 2y ago
I am reviewing papers for my substack regularly and there is a structure I believe works well that focuses on potential impact, structure, reading comprehension, github availability (where applicable), and problem relevance.
297.
by jsemrau 2y ago
"Our implementation is based on HuggingFace transformers where we register a new model opt-llama that supports the Layer-Condensed KV Cache." Not sure what this means? Would this work for a Mistral model as well?
298.
by jsemrau 2y ago
From a site like SEOroundtable, I would assume some kind of bias.
299.
by jsemrau 2y ago
>The size of the cached internal state of the network processing the book is much larger than the size of the book It's funny that sometimes people consider LLMs as compression engines. While a lot of information gets lost in each d
300.
by jsemrau 2y ago
Compared to Bing CoPilot, Google feels dated now. I know it sounds like a crazy thing to say, but being able to chat with your search results and refine what you are looking for. Get summarized explanations with sources is just wow.

More ›